风控领域下的表格学习

To provide rigorous uncertainty quantification for online learning models, we develop a framework for constructing uncertainty sets that provably control risk -- such as coverage of confidence intervals, false negative rate, or F1 score -- in the online setting. This extends conformal prediction to apply to a larger class of online learning problems. Our method guarantees risk control at any user-specified level even when the underlying data distribution shifts drastically, even adversarially, over time in an unknown fashion. The technique we propose is highly flexible as it can be applied with any base online learning algorithm (e.g., a deep neural network trained online), requiring minimal implementation effort and essentially zero additional computational cost. We further extend our approach to control multiple risks simultaneously, so the prediction sets we generate are valid for all given risks. To demonstrate the utility of our method, we conduct experiments on real-world tabular time-series data sets showing that the proposed method rigorously controls various natural risks. Furthermore, we show how to construct valid intervals for an online image-depth estimation problem that previous sequential calibration schemes cannot handle.

We introduce a framework for calibrating machine learning models so that their predictions satisfy explicit, finite-sample statistical guarantees. Our calibration algorithms work with any underlying model and (unknown) data-generating distribution and do not require model refitting. The framework addresses, among other examples, false discovery rate control in multi-label classification, intersection-over-union control in instance segmentation, and the simultaneous control of the type-1 error of outlier detection and confidence set coverage in classification or regression. Our main insight is to reframe the risk-control problem as multiple hypothesis testing, enabling techniques and mathematical arguments different from those in the previous literature. We use the framework to provide new calibration methods for several core machine learning tasks, with detailed worked examples in computer vision and tabular medical data.

The field of risk-constrained reinforcement learning (RCRL) has been developed to effectively reduce the likelihood of worst-case scenarios by explicitly handling risk-measure-based constraints. However, the nonlinearity of risk measures makes it challenging to achieve convergence and optimality. To overcome the difficulties posed by the nonlinearity, we propose a spectral risk measure-constrained RL algorithm, spectral-risk-constrained policy optimization (SRCPO), a bilevel optimization approach that utilizes the duality of spectral risk measures. In the bilevel optimization structure, the outer problem involves optimizing dual variables derived from the risk measures, while the inner problem involves finding an optimal policy given these dual variables. The proposed method, to the best of our knowledge, is the first to guarantee convergence to an optimum in the tabular setting. Furthermore, the proposed method has been evaluated on continuous control tasks and showed the best performance among other RCRL algorithms satisfying the constraints.

Average-reward Markov decision processes (MDPs) provide a foundational framework for sequential decision-making under uncertainty. However, average-reward MDPs have remained largely unexplored in reinforcement learning (RL) settings, with the majority of RL-based efforts having been allocated to discounted MDPs. In this work, we study a unique structural property of average-reward MDPs and utilize it to introduce Reward-Extended Differential (or RED) reinforcement learning: a novel RL framework that can be used to effectively and efficiently solve various learning objectives, or subtasks, simultaneously in the average-reward setting. We introduce a family of RED learning algorithms for prediction and control, including proven-convergent algorithms for the tabular case. We then showcase the power of these algorithms by demonstrating how they can be used to learn a policy that optimizes, for the first time, the well-known conditional value-at-risk (CVaR) risk measure in a fully-online manner, without the use of an explicit bi-level optimization scheme or an augmented state-space.

The quality and safety of grain and oil food are paramount to sustainable societal development and public health. Implementing early warning analysis and risk control is critical for the comprehensive identification and management of grain and oil food safety risks. However, traditional risk prediction models are limited by their inability to accurately analyze complex nonlinear data, while their reliance on centralized storage further undermines prediction credibility and traceability. This study proposes a deep learning risk prediction model integrated with a blockchain-based traceability mechanism. Firstly, a risk prediction model combining Grey Relational Analysis (GRA) and Bayesian-optimized Tabular Neural Network (TabNet-BO) is proposed, enabling precise and rapid fine-grained risk prediction of the data; Secondly, a risk prediction method combining blockchain and deep learning is proposed. This method first completes the prediction interaction with the deep learning model through a smart contract and then records the exceeding data and prediction results on the blockchain to ensure the authenticity and traceability of the data. At the same time, a storage optimization method is employed, where only the exceeding data is uploaded to the blockchain, while the non-exceeding data is encrypted and stored in the local database. Compared with existing models, the proposed model not only effectively enhances the prediction capability for grain and oil food quality and safety but also improves the transparency and credibility of data management.

Loan approval forecasting is an essential part of risk control based on finance, and its misjudgment can result in credit risks or unfair non-credit risks for borrowers. Manual heuristics do not scale, and popular machine learning approaches may lack fairness, calibration, and interpretability. In this paper, we present a fairness-aware framework for loan approval prediction that incorporates ensemble boosting models (LightGBM, XGBoost, AdaBoost, and CatBoost) and deep tabular architectures (TabNet and TabTransformer) to ensure fair, accurate, and regulator-compliant decision-making. In this, relational information is investigated, and an adaptive decision rule is learned under asymmetric costs using reinforcement learning. The pipeline includes fold-wise preprocessing to prevent information leakage; imbalanced decisions are addressed through SMOTE-Tomek and costsensitive weighting; and it has enforced monotonic constraints for domain similarity. The application of isotonic regression in probability calibration guarantees trustworthy outputs, and using reject inference can significantly reduce selection bias. All experiments and model evaluations were implemented on Google Colab, enabling reproducibility and scalable deployment testing. Extensive experiments on benchmark datasets demonstrate improved performance, with accuracy, specificity, F1-score, and ROC-AUC of 92.6%, 97.8%, 91.5%, and 96.9 %, respectively, consistently outperforming single models. Fairness-oriented modifications mitigated demographic imbalances and enhanced the interpretability of the model by providing both global and local explanations, utilizing SHAP and counterfactual explanations. Federated learning simulations validated federated learning's suitability for privacy-preserving cross-institution deployment. This work presents a comprehensive solution for intelligent loan approval automation that strikes a balance between predictive performance, fairness, and interpretability.

The use of Deep Neural Network (DNN) models in risk-based decision-making has attracted extensive attention with broad applications in medical, finance, manufacturing, and quality control. To mitigate prediction-related risks in decision making, prediction confidence or uncertainty should be assessed alongside the overall performance of algorithms. Recent studies on Bayesian deep learning helps quantify prediction uncertainty arises from input noises and model parameters. However, the normality assumption of input noise in these models limits their applicability to problems involving categorical and discrete feature variables in tabular datasets. In this paper, we propose a mathematical framework to quantify prediction uncertainty for DNN models. The prediction uncertainty arises from errors in predictors that follow some known finite discrete distribution. We then conducted a case study using the framework to predict treatment outcome for tuberculosis patients during their course of treatment. The results demonstrate under a certain level of risk, we can identify risk-sensitive cases, which are prone to be misclassified due to error in predictors. Comparing to the Monte Carlo dropout method, our proposed framework is more aware of misclassification cases. Our proposed framework for uncertainty quantification in deep learning can support risk-based decision making in applications when discrete errors in predictors are present.

Adaptive bitrate (ABR) algorithms must select per-chunk video quality under substantial network uncertainty. While reinforcement learning (RL) improves average Quality-of-Experience (QoE), trace-driven evaluations often reveal heavy-tailed stall events and brittle behavior under high-variance cellular links. This paper presents a risk-sensitive offline-RL ABR design that optimizes the lower tail of the return distribution via Conditional Value-at-Risk (CVaR) computed from a distributional Q-function. We conduct a full empirical evaluation using two public real-trace datasets: (i) 12 3G/HSDPA throughput logs from Norwegian mobile streaming sessions (UMass MMSys trace archive), and (ii) 20 4G/LTE bandwidth logs collected along routes in Ghent, Belgium (UGent/IDLab dataset). Using a Pensieve-style chunked streaming simulator and a standard QoE function (bitrate reward, rebuffer penalty, and smoothness penalty), we compare a buffer-based rule (BBA), robust model predictive control (RobustMPC), online tabular actor–critic (A2C), and an offline distributional RL method (Quantile Regression Conservative Q-Learning, QR-CQL) with a CVaR decision rule. Across 400 fixed test episodes on held-out traces, the risk-sensitive policy OfflineQR-CQL(CVaR@0.25) achieves mean QoE 104.91 (within 17.6% of the best policy, RobustMPC). Relative to online A2C, it improves mean QoE by -8.3% and reduces mean rebuffer time by -224.2%. Relative to RobustMPC, it improves mean QoE by -17.6% and reduces mean rebuffer time by -79.6%. Bucketed analysis by trace coefficient-of-variation shows the largest QoE gain in the highest-variability quartile (Q4), where OfflineQR-CQL(CVaR@0.25) exceeds RobustMPC by -27.59 QoE points. A CVaR sensitivity sweep confirms a controllable risk–reward trade-off governed by α.

Financial markets are dynamic, uncertain, and complex, requiring adaptive strategies to balance risk and return. Reinforcement learning (RL) offers new opportunities for modeling trading decisions, while traditional portfolio optimization and rule-based techniques remain popular. The relative strengths of benchmarks, rule-based filters, and RL models across investment horizons remain underexplored. This study fills the gap by comparing nine different strategies using daily data from 2,889 NASDAQ-listed stocks obtained from stockanalysis.com. Simulations were implemented in R (data preparation) and Python (modeling and RL algorithms). All agents started with 10,000 USD and were evaluated over three horizons: short-term (1 week), medium-term (1 month), and long-term (~160 trading days). The strategies included two benchmarks (Random Agent, Equal-Weight), two rule-based methods (Volatility-Momentum-Leadership, Mean-Reversion), three tabular RL models (Q-Learning, Safe Q-Learning, SARSA), and two deep RL algorithms (DQN, A2C).The findings show that the Volatility-Momentum-Leadership (VML) strategy achieved the best long-term performance, tripling the portfolio to 31,845. The Safe Q-Learning variant produced a more modest 25.2% gain, while the Mean-Reversion strategy collapsed under transaction costs. Deep RL models (e.g., DQN) performed well in the short term but declined over time. Nevertheless, rule-based filters delivered more consistent growth in the short horizon (1). The results also suggest that (2) RL shows promise mainly over short-horizon tactical trading but requires refinement for long-term stability, and (3) transaction costs are critical for strategy viability. This study contributes to portfolio optimization by showing that explainable rules provide robust baselines, while reinforcement learning may enhance timing and adaptability in practice.

Heart disease remains the leading cause of mortality worldwide, highlighting the need for accurate and interpretable risk prediction models. This study proposes Hybrid Tabular Network with Clinical meta-features and Quality-Control Mechanisms (HybridTabNet-QC), a hybrid deep learning framework designed to enhance clinical decision support systems. By combining a transformer-based tabular encoder with medically engineered meta-features, the model delivers interpretable, generalizable, and noise-robust predictions for heart disease risk. Unlike conventional deep learning models that treat features as raw inputs, HybridTabNet-QC fuses attention-based learning with domain-specific indicators, such as body mass index (BMI), pulse pressure, and cholesterol-glucose interaction scores that ensure predictions are both statistically sound and clinically grounded. The model was evaluated on two publicly available datasets, the Heart Disease dataset and the Cardiovascular Disease dataset from Kaggle. On the combined dataset, HybridTabNet-QC achieved 90.1% accuracy, 90.0% F1-score, and 93.6% AUC-ROC, outperforming traditional machine learning and standard deep learning baselines. The framework demonstrated strong robustness under feature noise and consistent generalization on external validation subsets. Interpretability analyses using LIME and SHAP confirmed that the model prioritizes clinically relevant features, supporting its suitability for real-world clinical decision support systems. These findings demonstrate the model's suitability for deployment in real-world healthcare environments where interpretability, robustness, and scalability are critical.

Tuberculosis (TB) remains a global health threat, ranking among the leading causes of mortality worldwide. In this context, machine learning (ML) has emerged as a transformative force, providing innovative solutions to the complexities associated with TB treatment.This study explores how machine learning, especially with tabular data, can be used to predict Tuberculosis (TB) treatment outcomes more accurately. It transforms this prediction task into a binary classification problem, generating risk scores from patient data sourced from NIKSHAY, India's national TB control program, which includes over 500,000 patient records. Data preprocessing is a critical component of the study, and the model achieved an recall of 98% and an AUC-ROC score of 0.95 on the validation set, which includes 20,000 patient records.We also explore the use of Natural Language Processing (NLP) for improved model learning. Our results, corroborated by various metrics and ablation studies, validate the effectiveness of our approach. The study concludes by discussing the potential ramifications of our research on TB eradication efforts and proposing potential avenues for future work. This study marks a significant stride in the battle against TB, showcasing the potential of machine learning in healthcare.

Food safety risk prediction is crucial for timely hazard detection and effective control. This study proposes a novel risk prediction method for food safety called TabNet-GRA, which combines a specialized deep learning architecture for tabular data (TabNet) with a grey relational analysis (GRA) to predict food safety risk. Initially, this study employed a GRA to derive comprehensive risk values from fused detection data. Subsequently, a food safety risk prediction model was constructed based on TabNet, and training was performed using the detection data as inputs and the comprehensive risk values calculated via the GRA as the expected outputs. Comparative experiments with six typical models demonstrated the superior fitting ability of the TabNet-based prediction model. Moreover, a food safety risk prediction and visualization system (FSRvis system) was designed and implemented based on TabNet-GRA to facilitate risk prediction and visual analysis. A case study in which our method was applied to a dataset of cooked meat products from a Chinese province further validated the effectiveness of the TabNet-GRA method and the FSRvis system. The method can be applied to targeted risk assessment, hazard identification, and early warning systems to strengthen decision making and safeguard public health by proactively addressing food safety risks.

Uncertainty remains a critical challenge in healthcare AI, since predictive errors can directly compromise patient safety and undermine trust. Structured clinical datasets in healthcare are frequently characterized by heterogeneous acquisition protocols, incomplete records, and inconsistent or noisy encodings. This inflates aleatoric uncertainty and weakens calibration. These challenges are exemplified in lung cancer risk modeling, where small cohorts, variable collection practices, and limited feature quality make the problem especially acute. Significant advances in uncertainty quantification (UQ) have been achieved in imaging and signal processing through Bayesian inference, evidential learning, and robust architectural designs. In contrast, tabular clinical datasets remain a critical yet underexplored domain. Addressing this gap requires methods that are lightweight, certifiable, and effective on noisy datasets without relying on large models or data. Considering this challenges, we propose a frequency-aware hybrid representation that combines Principal Component Analysis (PCA) with the Discrete Cosine Transform (DCT). Using mutual information (MI)–based feature ordering, the framework suppresses high-frequency artifacts while preserving discriminative structure. As the framework was applied to a publicly available lung cancer dataset, it demonstrated an accuracy improvement from 98.1% to 99.7%, reduced Negative Log-Likelihood (NLL) by 82% from 5.25% to 0.94%, lowered aleatoric uncertainty from 10.50% to 3.35% (68% reduction), and preserved AUROC at 99%. We evaluated the framework across three publicly available lung cancer datasets where it demonstrated a reduction in aleatoric uncertainty by 7% on an average, confirming generalizability. The Wilcoxon signed-rank test confirms that the results are statistically significant. This work shows that part of the ‘irreducible’ variability is actually compressible noise, thereby facilitating more reliable and uncertainty-aware AI for healthcare.

Reinforcement Learning (RL) is crucial for data-driven decision-making but suffers from sample inefficiency. This poses a risk to system safety and can be costly in real-world environments with physical interactions. This paper proposes a human-inspired framework to improve the sample efficiency of RL algorithms, which gradually provides the learning agent with simpler but similar tasks that progress toward the main task. The proposed method does not require pre-training and can be applied to any goal, environment, and RL algorithm, including value-based and policy-based methods, as well as tabular and deep-RL methods. The framework is evaluated on a Random Walk and optimal control problem with constraint, showing good performance in improving the sample efficiency of RL-learning algorithms.

Rationale and objectives We evaluate the automatic identification of type 2 diabetes from neck-to-knee, two-point Dixon MRI scans with 3D convolutional neural networks on a large, population-based dataset. To this end, we assess the best combination of MRI contrasts and stations for diabetes prediction, and the benefit of integrating risk factors. Materials and methods Subjects with type 2 diabetes mellitus have been identified in the prospective UK Biobank Imaging study, and a matched control sample has been created to avoid confounding bias. Five-fold cross-validation is used for the evaluation. All scans from the two-point Dixon neck-to-knee sequence have been standardized. A neural network that considers multi-channel MRI input was developed and integrates clinical information in tabular format. An ensemble strategy is used to combine multi-station MRI predictions. A subset with quantitative fat measurements is identified for comparison to prior approaches. Results MRI scans from 3406 subjects (mean age, 66.2 years ± 7.1 [standard deviation]; 1128 women) were analyzed with 1703 diabetics. A balanced accuracy of 78.7 %, AUC ROC of 0.872, and an average precision of 0.878 was obtained for the classification of diabetes. The ensemble over multiple Dixon MRI stations yields better performance than selecting the individually best station. Moreover, combining fat and water scans as multi-channel inputs to the networks improves upon just using single contrasts as input. Integrating clinical information about known risk factors of diabetes in the network boosts the performance across all stations and the ensemble. The neural network achieved superior results compared to the prediction based on quantitative MRI measurements. Conclusions The developed deep learning model accurately predicted type 2 diabetes from neck-to-knee two-point Dixon MRI scans.

Automated stock trading is now the de-facto way that investors have chosen to obtain high profits in the stock market while keeping risk under control. One of the approaches is to create agents employing Reinforcement Learning (RL) algorithms to learn and decide whether or not to operate in the market in order to achieve maximum profit. Automated financial trading systems can learn how to trade optimally while interacting with the market pretty much like a human investor learns how to trade. In this research, a simple RL agent was implemented using the SARSA algorithm. Next, it was tested against 10 stocks from Brazilian stock market B3 (Bolsa, Brasil, Balcão). Results from experiments showed that the agent was able to provide high profits with less risk when compared to a supervised learning agent that used a LSTM neural network.

Loan default prediction is a core task in financial risk management, where traditional models face challenges of data imbalance, complex features, and interpretability limitations. This study proposes a hybrid CTGAN-MLP model, which employs the Conditional Tabular Generative Adversarial Network (CTGAN) to generate balanced synthetic data and leverages the nonlinear model capability of the Multilayer Perceptron (MLP) to improve prediction accuracy. Experiments compare the classification performance of MLP with traditional machine learning models and evaluate the impact of dataset balancing. Results show that the CTGAN-MLP model significantly enhances default identification, particularly in classifying minority samples. This research provides a novel approach integrating data augmentation and intelligent model for financial risk control, offering both theoretical innovation and practical guidance.

The article is devoted to the analysis of modern approaches to forecasting financial trends using artificial intelligence (AI) in the conditions of 2024-2025, when the role of news factors, the speed of information dissemination and natural language processing methods has sharply increased. Key classes of models (machine learning on tabular features, neural network models of time series, transformers, ensembles) and practices for extracting signals from news streams are considered. Special attention is paid to the use of large language models (LLM) and Retrieval-Augmented Generation (RAG) approaches for structuring news, extracting events and building features, as well as the limitations of such solutions: data drift, retraining, interpretation difficulties, risks of LLM “hallucinations” and the need for compliance control. A step-by-step scheme for developing a software solution to support analytics and risk management is proposed: data collection and validation, feature formation, training in walk-forward modes, quality and stability assessment, implementation with MLOps monitoring and routine updates. It is concluded that the greatest applied effect is achieved when using AI as a decision support tool, when forecasts are accompanied by reliability metrics, explanations and scenario analysis.

— Cardiovascular disease (CVD) is a globally significant health issue that presents with a multitude of risk factors and complex physiology, making early detection, avoidance, and effective management a challenge. Early detection is essential for effective treatment of CVD, and typical approaches involve an integrated strategy that includes lifestyle modifications like exercise and diet, medications to control risk factors like high blood pressure and cholesterol, interventions like angioplasties or bypass surgery in extreme cases, and ongoing surveillance to prevent complications and promote heart function. Traditional approaches often rely on manual interpretation, which is time-consuming and prone to error. In this paper, proposed study uses an automated detection method using machine learning. The CNN and XGBoost algorithms' greatest characteristics are combined in the hybrid technique. CNN is excellent in identifying pertinent features from medical images, while XGBoost performs well with tabular data. By including these strategies, the model's robustness and precision in predicting CVD are both increased. Furthermore, data normalization techniques are employed to confirm the accuracy and consistency of the model's projections. By standardizing the input data, the normalization procedure lowers variability and increases the model's ability to extrapolate across instances. This work explores a novel approach to CVD detection using a CNN/XGBoost hybrid model. The hybrid CNN-XGBoost and explainable AI system has undergone extensive testing and validation, and its performance in accurately detecting CVD is encouraging. Due to its ease of use and effectiveness, this technique may be applied in clinical settings, potentially assisting medical professionals in the prompt assessment and care of patients with cardiovascular disease.

Given the huge volume of cross-border flows, effective and efficient control of trade becomes more crucial in protecting people and society from illicit trade. However, limited accessibility of the transaction-level trade datasets hinders the progress of open research, and lots of customs administrations have not benefited from the recent progress in data-based risk management. In this paper, we introduce an import declaration dataset to facilitate the collaboration between domain experts in customs administrations and researchers from diverse domains, such as data science and machine learning. The dataset contains 54,000 artificially generated trades with 22 key attributes, and it is synthesized with conditional tabular GAN while maintaining correlated features. Synthetic data has several advantages. First, releasing the dataset is free from restrictions that do not allow disclosing the original import data. The fabrication step minimizes the possible identity risk which may exist in trade statistics. Second, the published data follow a similar distribution to the source data so that it can be used in various downstream tasks. Hence, our dataset can be used as a benchmark for testing the performance of any classification algorithm. With the provision of data and its generation process, we open baseline codes for fraud detection tasks, as we empirically show that more advanced algorithms can better detect fraud.

Few-shot tabular learning, in which machine learning models are trained with a limited amount of labeled data, provides a cost-effective approach to addressing real-world challenges. The advent of Large Language Models (LLMs) has sparked interest in leveraging their pre-trained knowledge for few-shot tabular learning. Despite promising results, existing approaches either rely on test-time knowledge extraction, which introduces undesirable latency, or text-level knowledge, which leads to unreliable feature engineering. To overcome these limitations, we propose Latte, a training-time knowledge extraction framework that transfers the latent prior knowledge within LLMs to optimize a more generalized downstream model. Latte enables general knowledge-guided downstream tabular learning, facilitating the weighted fusion of information across different feature values while reducing the risk of overfitting to limited labeled data. Furthermore, Latte is compatible with existing unsupervised pre-training paradigms and effectively utilizes available unlabeled samples to overcome the performance limitations imposed by an extremely small labeled dataset. Extensive experiments on various few-shot tabular learning benchmarks demonstrate the superior performance of Latte, establishing it as a state-of-the-art approach in this domain. Our code is available at https://github.com/ruxueshi/Latte.git.

With the rapid development of Internet technology, Internet finance has entered thousands of households, bringing a lot of convenience to people’s lives. As a financial model based on Internet technology, compared with traditional bank loans, online loans have lower operating costs and faster returns, so they are developing rapidly. Providing users with better and faster services while standardizing operations has always been the development goal of various Internet finance companies. At present, many domestic Internet financial enterprises are facing many problems such as difficulty in risk control. Therefore, this paper makes full use of deep learning algorithms to build an Internet financial risk control system. After in-depth analysis and research on the deep learning algorithm, the Internet financial risk control system is divided into several modules. The project mainly includes model management module, user behavior analysis module, alarm management module, monitoring module, product management module, etc., and then by analyzing the test results of each test scenario, it is concluded that the performance test of the system design meets the actual needs of users, and simulates the test. It is carried out in accordance with the constraints and regulations of the test plan, and the performance test meets the standard. The system not only ensures the safety of the company's funds, but also helps the company to form a smooth and effective financial risk control process. This paper designs a class of effective management systems by applying deep learning algorithms to the field of Internet financial risk control.

BACKGROUND AND OBJECTIVE Modeling structured medical tabular data presents significant challenges due to complex sample dependencies and non-linear feature interactions. Existing methods, which primarily focus on single-disease prediction, often exhibit limited capability in forecasting critical progression in patients with multimorbidity. To address this, we propose GATET, a novel architecture that integrates graph neural networks, deep tabular learning, and population subgraph partitioning to improve predictive accuracy for multimorbid patients. METHODS GATET comprises three core modules: (1) Dependency Feature Extraction (DFE), which generates trainable adjacency matrices guided by medical prior knowledge; (2) Attentive Aggregation for Constructing Graphs (CGsA), which employs dual-channel graph attention networks to capture intricate relationships within the population graph, and (3) Feature Weighting based on TabNet (FWT), which preserves TabNet's interpretability while removing its global modeling mechanism to eliminate redundant computations. The implementation is publicly available at https://www.researchgate.net/profile/Chengjie-Li-7. RESULTS Extensive repeated experiments with statistical hypothesis testing, performed on clinical data from a tertiary hospital in Southwest China, demonstrate that GATET improves prediction accuracy by approximately 10% over baseline models and achieves superior performance across additional metrics. Domain adaptation experiments on multiple datasets confirm its effectiveness for other disease prediction tasks. Supplementary analyses, including parameter sensitivity studies and graph-aggregated feature selection, empirically validate the importance of age-based stratification in multimorbid populations. CONCLUSIONS Comprehensive comparative evaluations highlight GATET's strong potential for predicting critical disease progression in multimorbid patients. This work presents an effective strategy for integrating prior medical knowledge into graph-based frameworks, advancing predictive analytics for structured tabular data and delivering tangible improvements for complex clinical prediction.

OBJECTIVE Machine learning (ML) models have been extensively used for tabular data classification but recent works have been developed to transform tabular data into images, aiming to leverage the predictive performance of convolutional neural networks (CNNs). However, most of these approaches fail to convert data with a low number of samples and mixed-type features. This study aims: to evaluate the performance of the tabular-to-image method named low mixed-image generator for tabular data (LM-IGTD); and to assess the effectiveness of transfer learning and fine-tuning for improving predictions on tabular data. METHODS We employed two public tabular datasets with patients diagnosed with cardiovascular diseases (CVDs): Framingham and Steno. First, both datasets were transformed into images using LM-IGTD. Then, Framingham, which contains a larger set of samples than Steno, is used to train CNN-based models. Finally, we performed transfer learning and fine-tuning using the pre-trained CNN on the Steno dataset to predict CVD risk. RESULTS The CNN-based model with transfer learning achieved the highest AUCORC in Steno (0.855), outperforming ML models such as decision trees, K-nearest neighbours, least absolute shrinkage and selection operator (LASSO) support vector machine and TabPFN. This approach improved accuracy by 2% over the best-performing traditional model, TabPFN. CONCLUSION To the best of our knowledge, this is the first study that evaluates the effectiveness of applying transfer learning and fine-tuning to tabular data using tabular-to-image approaches. Through the use of CNNs' predictive capabilities, our work also advances the diagnosis of CVD by providing a framework for early clinical intervention and decision-making support.

Federated Learning (FL) has emerged as a potentially powerful privacy-preserving machine learning method-ology, since it avoids exchanging data between participants, but instead exchanges model parameters. FL has traditionally been applied to image, voice and similar data, but recently it has started to draw attention from domains including financial services where the data is predominantly tabular. However, the work on tabular data has not yet considered potential attacks, in particular attacks using Generative Adversarial Networks (GANs), which have been successfully applied to FL for non-tabular data. This paper is the first to explore leakage of private data in Federated Learning systems that process tabular data. We design a Generative Adversarial Networks (GANs)-based attack model which can be deployed on a malicious client to reconstruct data and its properties from other participants. As a side-effect of considering tabular data, we are able to statistically assess the efficacy of the attack (without relying on human observation such as done for FL for images). We implement our attack model in a recently developed generic FL software framework for tabular data processing. The experimental results demonstrate the effectiveness of the proposed attack model, thus suggesting that further research is required to counter GAN-based privacy attacks.

Tabular data -- structured, heterogeneous, spreadsheet-style data with rows and columns -- is widely used in practice across many domains. However, while recent foundation models have reduced the need for developing task-specific datasets and predictors in domains such as language modeling and computer vision, this transfer learning paradigm has not had similar impact in the tabular domain. In this work, we seek to narrow this gap and present TabuLa-8B, a language model for tabular prediction. We define a process for extracting a large, high-quality training dataset from the TabLib corpus, proposing methods for tabular data filtering and quality control. Using the resulting dataset, which comprises over 2.1B rows from over 4M unique tables, we fine-tune a Llama 3-8B large language model (LLM) for tabular data prediction (classification and binned regression) using a novel packing and attention scheme for tabular prediction. Through evaluation across a test suite of 329 datasets, we find that TabuLa-8B has zero-shot accuracy on unseen tables that is over 15 percentage points (pp) higher than random guessing, a feat that is not possible with existing state-of-the-art tabular prediction models (e.g. XGBoost, TabPFN). In the few-shot setting (1-32 shots), without any fine-tuning on the target datasets, TabuLa-8B is 5-15 pp more accurate than XGBoost and TabPFN models that are explicitly trained on equal, or even up to 16x more data. We release our model, code, and data along with the publication of this paper.

With the development of Internet finance, the importance of loan risk control is increasingly manifested. Risk control is the core part of traditional financial industry and Internet finance. After investigating the latest developments in credit risk control algorithms, an improved stacking integrated learning algorithm is proposed. By improving the feature selection steps, and using 5 different learners for stacking integration, the performance of the model is improved. The basic learners used include: Logistic Regression, Random Forest, GBDT, XGBoost, LightGBM, among which there are both strong learners and weak learners. Compared with traditional integrated learning methods, the accuracy of strong learners can be fully utilized, and use weak learners to reduce overfitting. Finally, the accuracy and generalization performance of the model are improved.

Automated indoor environmental control is a research topic that is beginning to receive much attention in smart home automation. All machine learning models proposed to date for this purpose have relied on reinforcement learning using simple metrics of comfort as reward signals. Unfortunately, such indicators do not take into account individual preferences and other elements of human perception. This research explores an alternative (albeit closely related) paradigm called imitation learning. In the proposed architecture, machine learning models are trained with tabular data pertaining to environmental control activities of the real occupants of a residential unit. This eliminates the need for metrics that explicitly quantify human perception of comfort. Moreover, this article introduces the recently proposed deep attentive tabular neural network (TabNet) into smart home research by incorporating TabNet-based components within its overall framework. TabNet has consistently outperformed all other popular machine learning models in a variety of other application domains, including gradient boosting, which was previously considered ideal for learning from tabular data. The results obtained herein strongly suggest that TabNet is the best choice for smart home applications. Simulations conducted using the proposed architecture demonstrate its effectiveness in reproducing the activity patterns of the home unit’s actual occupants.

This study presents an AI-driven risk control framework aimed at effectively managing the risks associated with health insurance fund operations. The backdrop reveals that the increase in fraudulent activities has contributed to a significant slowdown in the growth rate of health insurance fund income compared to expenditures in China, leading to a decline in surplus rates. To address this challenge, the proposed framework integrates unsupervised and supervised learning methodologies for risk identification and quantification. Specifically, we employ Gaussian Mixture Models (GMM) for clustering medical behaviors to detect anomalies, followed by the application of the LightGBM model for risk classification and quantification. Experimental results demonstrate the framework’s robust capabilities in identifying potential fraud, underscoring that frequent medical visits and significant expenditures on non-essential medications are key indicators of fraudulent behavior. In conclusion, the proposed framework not only enhances the transparency and efficiency of health insurance fund management but also provides a solid foundation for implementing effective risk control measures.

Young motorcyclists, particularly those aged 15–24, face a heightened risk of severe crashes due to factors such as speeding, traffic violations, and helmet non-use. This study aims to identify key factors influencing crash severity by analyzing 10,726 young motorcyclist crashes in Texas from 2017 to 2022. Two advanced tabular deep learning models, ARM-Net and MambaNet, were employed, using an advanced resampling technique to address class imbalance. The models were trained to classify crashes into three severity levels: Fatal/Severe (KA), Moderate/Minor (BC), and No Injury (O). ARM-Net achieved an accuracy of 87%, outperforming MambaNet's 86%, with both models excelling in predicting severe and no-injury crashes while facing challenges in moderate crash classification. Key findings highlight the significant influence of demographic, environmental, and behavioral factors on crash outcomes. The study underscores the need for targeted interventions, including stricter helmet enforcement and educational programs customized to young motorcyclists. These insights provide valuable guidance for policymakers in developing evidence-based strategies to enhance motorcyclist safety and reduce crash severity.

No abstract available

In the evolving landscape of financial technology, credit scoring systems must adapt to increasingly complex data environments to ensure accurate and fair lending decisions. Traditional credit models rely heavily on structured data such as income, credit history, and debt ratios, but often fall short in assessing borrowers with limited credit footprints or non-traditional financial behaviors. This study proposes a hybrid deep neural network (DNN) model that integrates structured financial indicators with unstructured textual data—including customer service interactions, financial news, and social media sentiment—to enhance credit risk prediction. We collected and preprocessed a multimodal dataset comprising over 100,000 loan profiles, developed a bi-directional LSTM architecture for text processing, and fused it with structured data via a deep learning framework. Our model was evaluated against benchmark algorithms including logistic regression, random forest, XGBoost, and single-input DNNs. Experimental results show that the hybrid DNN significantly outperforms traditional models, achieving an accuracy of 87% and an AUC-ROC of 0.91. These findings underscore the potential of multimodal deep learning in transforming credit scoring systems, improving model precision, and expanding financial inclusion. The proposed model offers a scalable and robust framework for future credit evaluation tools in data-rich financial ecosystems.

The Basel Accord emphasizes the necessity of employing internal data models to manage key credit risk components, including Probability of Default (PD), Loss Given Default (LGD), and Exposure At Default (EAD). Among these, internal datasets are critical for estimating PD, a fundamental measure of borrower creditworthiness. Nevertheless, practical application often faces challenges due to incomplete datasets, which can skew analyses and undermine the accuracy of credit scoring models. Traditional approaches to addressing missing data, such as sample deletion or mean imputation, are widely used; however, they often prove insufficient for accurate prediction. Consequently, imputation methods are typically favored over deletion, as they allow for the full utilization of available data. Recent advancements have introduced more sophisticated techniques, such as Generative Adversarial Imputation Networks (GAIN), which utilize a generative adversarial network to model data distributions and impute missing values with greater precision than conventional methods. Building on these developments, this study proposes a novel imputation approach, SMART (Structured Missingness Analysis and Reconstruction Technique) for credit scoring datasets. SMART consists of two primary stages: first, it normalizes and denoises the dataset using randomized Singular Value Decomposition (rSVD), followed by the implementation of GAIN to impute missing values. Experimental results demonstrate that SMART significantly outperforms existing state-of-the-art methods, particularly in high missing data contexts (20%, 50%, and 80%), with improvements in imputation accuracy of 7.04%, 6.34%, and 13.38%, respectively. In conclusion, SMART represents a substantial advancement in handling incomplete credit scoring datasets, leading to more precise PD estimation and enhancing the robustness of credit risk management models.

With the rapid development of big data and machine learning technologies, the application of sentiment analysis in financial credit scoring is gradually revolutionizing the traditional risk assessment paradigm. Traditional credit scoring models rely on structured financial data, making it difficult to capture the dynamic sentimental signals of borrowers and the potential impact of market sentiment. In this paper, this paper systematically explore the practical path of sentiment analysis technology to build a multi-dimensional credit risk portrait by mining unstructured data such as social media comments, news texts, customer feedback, etc., combined with machine learning algorithms (e.g., BERT, XGBoost, LSTM). Studies have shown that sentiment features can significantly improve the predictive ability of credit scoring models: in terms of technology integration, big data infrastructure (e.g., Hadoop, Spark) supports the real-time processing of massive text, while deep learning models (e.g., BLIP-NLP) realize the fine-grained extraction of sentiment signals. However, sentiment analysis in the financial domain still faces challenges such as data noise, dynamic adaptation, model interpretability, and privacy ethics.

A multi-criteria optimization mathematical model of credit scoring is proposed. The model is derived using a nonlinear trade-off scheme to solve multi-criteria optimization problems, allowing for the construction of a Pareto-optimal solution. The proposed approach forms an integrated assessment of a borrower’s creditworthiness based on a structured set of indicators that reflect the financial, credit, and social profile of clients. The model is designed for use in intelligent CRM and ERP systems operating on Big Data and does not rely on labeled training samples, making it applicable to unsupervised learning tasks. It can also serve as a foundational layer for further deep-learning analysis. Methodological steps for implementing the model, from indicator normalization to final decision-making, are described. A technological implementation demonstrates the model’s effectiveness in automated loan decisions and fraud detection.

In emerging markets like Vietnam, where student borrowers often lack traditional credit histories, accurately predicting loan eligibility remains a critical yet underexplored challenge. While machine learning and deep learning techniques have shown promise in credit scoring, their comparative performance in the context of student loans has not been thoroughly investigated. This study aims to evaluate and compare the predictive effectiveness of four supervised learning models—such as Random Forest, Gradient Boosting, Support Vector Machine, and Deep Neural Network (implemented with PyTorch version 2.6.0)—in forecasting student credit eligibility. Primary data were collected from 1024 university students through structured surveys covering academic, financial, and personal variables. The models were trained and tested on the same dataset and evaluated using a comprehensive set of classification and regression metrics. The findings reveal that each model exhibits distinct strengths. Deep Learning achieved the highest classification accuracy (85.55%), while random forest demonstrated robust performance, particularly in providing balanced results across classification metrics. Gradient Boosting was effective in recall-oriented tasks, and support vector machine demonstrated strong precision for the positive class, although its recall was lower compared to other models. The study highlights the importance of aligning model selection with specific application goals, such as prioritizing accuracy, recall, or interpretability. It offers practical implications for financial institutions and universities in developing machine learning and deep learning tools for student loan eligibility prediction. Future research should consider longitudinal data, behavioral factors, and hybrid modeling approaches to further optimize predictive performance in educational finance.

The increasing complexity and volume of financial and behavioral data in modern credit scoring and risk assessment present significant challenges to traditional modeling methods. Existing approaches often struggle with integrating structured numerical records and unstructured user behavior signals, limiting their ability to capture meaningful temporal and non-linear patterns. In the swiftly transforming domain of computational science, the incorporation of sophisticated machine learning algorithms has emerged as a critical driver in addressing these challenges. Although traditional statistical approaches provide foundational value, they frequently fall short when faced with the task of capturing the nuanced, non-linear associations embedded within extensive datasets, thereby limiting predictive precision. To address these shortcomings, we propose an innovative deep learning paradigm that effectively integrates structured financial information with unstructured behavioral data, thereby bolstering the reliability and accuracy of predictive analytics. Our approach features a thorough feature engineering pipeline that includes statistical indicators, temporal trends, and aggregated financial attributes—aimed at constructing a comprehensive representation of the data environment. At the core of our framework is a hybrid neural network architecture, which leverages Long Short-Term Memory (LSTM) units to handle sequential dependencies alongside dense layers that model complex interactions among features. This configuration enables the simultaneous learning of both temporal dynamics and high-level abstractions. To prevent overfitting and to promote better generalization, we incorporate adaptive regularization strategies that adjust penalization levels in response to validation metrics. We confront the issue of class imbalance by applying dynamic re-sampling and weight adjustment techniques, ensuring balanced model performance across varied data segments. Extensive evaluations on standard datasets validate the effectiveness of our proposed model, revealing enhanced prediction accuracy and model interpretability when compared to more traditional techniques. These results highlight the promise of advanced deep learning integration in computational science applications, offering a pathway toward more insightful and dependable predictive solutions.

Credit scoring is a high-stakes task in financial services, where model decisions directly impact individuals'access to credit and are subject to strict regulatory scrutiny. While Quantum Machine Learning (QML) offers new computational capabilities, its black-box nature poses challenges for adoption in domains that demand transparency and trust. In this work, we present IQNN-CS, an interpretable quantum neural network framework designed for multiclass credit risk classification. The architecture combines a variational QNN with a suite of post-hoc explanation techniques tailored for structured data. To address the lack of structured interpretability in QML, we introduce Inter-Class Attribution Alignment (ICAA), a novel metric that quantifies attribution divergence across predicted classes, revealing how the model distinguishes between credit risk categories. Evaluated on two real-world credit datasets, IQNN-CS demonstrates stable training dynamics, competitive predictive performance, and enhanced interpretability. Our results highlight a practical path toward transparent and accountable QML models for financial decision-making.

Algorithmic bias in artificial intelligence (AI) systems has raised significant ethical concerns, particularly in critical applications such as credit scoring, where fairness and accuracy are paramount. This study proposes a novel framework that integrates Convolutional Neural Networks (CNN) with the Synthetic Minority Oversampling Technique (SMOTE) to address data imbalance and mitigate algorithmic bias. The approach leverages CNN's ability to capture complex nonlinear relationships within structured credit data while employing SMOTE to generate synthetic samples for underrepresented classes, ensuring a balanced training dataset. By incorporating fairness-aware metrics and optimization strategies, the proposed framework not only improves predictive accuracy but also promotes equitable decision-making. Experimental evaluations on real-world credit scoring datasets demonstrate that this hybrid method outperforms traditional models, achieving higher classification performance while reducing disparities across demographic groups. This research highlights the potential of combining deep learning and oversampling techniques to build fairer and more transparent AI systems, paving the way for ethical advancements in financial decision-making.

Small and Medium-sized Enterprises (SMEs) are known to play a vital role in economic growth, employment, and innovation. However, they tend to face significant challenges in accessing credit due to limited financial histories, collateral constraints, and exposure to macroeconomic shocks. These challenges make an accurate credit risk assessment by lenders crucial, particularly since SMEs frequently operate within interconnected firm networks through which default risk can propagate. This paper presents and tests a novel approach for modelling the risk of SME credit, using a unique large data set of SME loans provided by a prominent financial institution. Specifically, our approach employs Graph Neural Networks to predict SME default using multilayer network data derived from common ownership and financial transactions between firms. We show that combining this information with traditional structured data not only improves application scoring performance, but also explicitly models contagion risk between companies. Further analysis shows how the directionality and intensity of these connections influence financial risk contagion, offering a deeper understanding of the underlying processes. Our findings highlight the predictive power of network data, as well as the role of supply chain networks in exposing SMEs to correlated default risk.

We introduce Omega^2, a Large Language Model-driven framework for corporate credit scoring that combines structured financial data with advanced machine learning to improve predictive reliability and interpretability. Our study evaluates Omega^2 on a multi-agency dataset of 7,800 corporate credit ratings drawn from Moody's, Standard&Poor's, Fitch, and Egan-Jones, each containing detailed firm-level financial indicators such as leverage, profitability, and liquidity ratios. The system integrates CatBoost, LightGBM, and XGBoost models optimized through Bayesian search under temporal validation to ensure forward-looking and reproducible results. Omega^2 achieved a mean test AUC above 0.93 across agencies, confirming its ability to generalize across rating systems and maintain temporal consistency. These results show that combining language-based reasoning with quantitative learning creates a transparent and institution-grade foundation for reliable corporate credit-risk assessment.

P2P online lending is a vital component of the fintech sector, and its rapid growth has heightened demands for predicting default risks. Traditional credit scoring models primarily rely on structured metrics such as FICO scores and debt-to-income ratios (DTI), yet they often overlook the predictive value of soft information like loan purpose. Using Lending Club’s 2016 loan data as a case study, this research constructs comparative models and employs machine learning methods including decision trees, random forests, logistic regression, and XGBoost to empirically examine the impact of loan purpose on default prediction accuracy. Results reveal a significant correlation between loan purpose and default risk, demonstrating its positive role in enhancing model accuracy. Particularly under recall-prioritized risk control strategies, the logistic regression and XGBoost models exhibit superior performance in default identification coverage. This study not only enriches the theoretical dimensions of credit risk modeling but also provides empirical evidence and practical pathways for P2P platforms and credit institutions to optimize their risk control systems.

assess the creditworthiness of both individuals and businesses. Evaluating the risk of business failure is especially significant for stakeholders like lenders and investors. Credit scoring provides a structured and data-driven method to predict these risks by analyzing financial, operational, and historical information. Applications of credit scoring include risk assessment, financial stability forecasting, trend identification, risk-based pricing, and default prediction. By providing a data-driven evaluation of credit risk, it enables institutions to make informed decisions, reduce potential losses, and improve risk management strategies. This research aims to bridge this gap by analyzing the effectiveness of neural network ensembles and hybrid neural network models using three standard credit scoring benchmark datasets: Australian, German, and Japanese. Experimental results show that while standalone neural networks achieve accuracies of 87.44%, 83.37%, and 85.08% respectively, ensemble models (weighted voting) improve performance to 92.75%, 89.34%, and 89.97%. Hybrid neural networks outperform both in the Australian dataset (93.61%), but show similar performance in the German (89.45%) and Japanese (89.17%) datasets. Although hybrid models demonstrate slightly higher accuracy on one dataset, the overall difference between hybrid and ensemble models is not statistically significant. This study provides a comprehensive comparative analysis to support the development of more accurate bankruptcy prediction systems and credit risk modeling strategies.

The advent of artificial intelligence has significantly enhanced credit scoring technologies. Despite the remarkable efficacy of advanced deep learning models, mainstream adoption continues to favor tree-structured models due to their robust predictive performance on tabular data. Although pretrained models have seen considerable development, their application within the financial realm predominantly revolves around question-answering tasks and the use of such models for tabular-structured credit scoring datasets remains largely unexplored. Tabular-oriented large models, such as TabPFN, has made the application of large models in credit scoring feasible, albeit can only processing with limited sample sizes. This paper provides a novel framework to combine tabular-tailored dataset distillation technique with the pretrained model, empowers the scalability for TabPFN. Furthermore, though class imbalance distribution is the common nature in financial datasets, its influence during dataset distillation has not been explored. We thus integrate the imbalance-aware techniques during dataset distillation, resulting in improved performance in financial datasets (e.g., a 2.5% enhancement in AUC). This study presents a novel framework for scaling up the application of large pretrained models on financial tabular datasets and offers a comparative analysis of the influence of class imbalance on the dataset distillation process. We believe this approach can broaden the applications and downstream tasks of large models in the financial domain.

Credit scoring is pivotal in financial institutions' risk management. This research integrates XG Boost and neural networks to develop a robust credit scoring model. XG Boost handles structured data and complex relationships, while neural networks excel in learning intricate patterns. The study aims to enhance model interpretability and accuracy. Evaluation using comprehensive metrics reveals the model's effectiveness. The integrated framework achieves 89% accuracy, outperforming traditional approaches. This study highlights the enhanced predictive power of the XG Boost and neural network model. It offers a promising solution to address the limitations of existing credit scoring methods. By leveraging the complementary strengths of both techniques, the model captures diverse features and nuances inherent in credit data. Ultimately, this research contributes to improving decision-making processes in financial institutions, promoting transparency and accuracy in credit scoring.

Credit Score plays a role in various fields, especially where it requires a lot of data, such as total income, credit history and can also make decisions in Google Analytics. It can represent a mathematical model which are based on statistical methods and can be used with a large amount of information. Various methods are used to identify an efficient score. It plays a crucial role in financial decision making, which impacts institutions and customer satisfaction. Traditional credit scoring models mostly rely on statistical methods and structured data, which sometimes fail to capture complex datasets. In this research, we are using an ensemble deep learning approach that will enhance the score of the credit by converting the data into heatmap images, which we will use as the representation of our data. We will use Principal Component Analysis (PCA) for dimension reduction, alongside Convolutional Neural Networks (CNNs), which will help in feature extraction from heatmaps. For extracting CNN features, we will use two models: Bagging Regressor and XGBoost Regressor. Our findings from the dataset it will help to demonstrate that heatmap-based learning models will improve score accuracy and provide a novel approach for financial assessment.

In an era of rapidly evolving credit landscapes, accurately predicting loan default risks has become paramount for the financial stability and profitability of lending institutions. Traditional credit scoring methods, which rely primarily on structured financial data such as income statements, credit history, and repayment records, are increasingly proving inadequate in capturing the full risk profile of modern borrowers. The inclusion of unstructured data—ranging from transaction narratives, social media behavior, to customer service interactions—offers a promising new dimension for enhancing credit risk modeling. This paper presents an integrated approach to developing predictive models for via loan default risks by combining structured and unstructured financial data across multiple lending institutions. It begins by categorizing the types of available data sources, detailing data preprocessing techniques for unstructured inputs using natural language processing (NLP), and outlining key feature engineering steps. The paper evaluates the effectiveness of machine learning algorithms, including gradient boosting machines, support vector machines, and deep neural networks, in predicting default probabilities with higher granularity. Furthermore, the study emphasizes the importance of inter-institutional data aggregation and model generalizability, addressing issues related to data privacy, regulatory compliance (e.g., GDPR, CCPA), and fairness in automated decision-making. It proposes a federated learning framework to allow collaborative model training without compromising sensitive customer data. Case studies demonstrate significant improvements in prediction accuracy and early warning lead times when combining hybrid data streams. Ultimately, this research underscores the critical role of data diversity and algorithmic sophistication in mitigating credit risk and enhancing portfolio resilience in the lending sector.

In this paper, the credit scoring problem is studied by incorporating networked information, where the advantages of such incorporation are investigated theoretically in two scenarios. Firstly, a Bayesian optimal filter is proposed to provide risk prediction for lenders assuming that published credit scores are estimated merely from structured financial data. Such prediction can then be used as a monitoring indicator for the risk management in lenders' future decisions. Secondly, a recursive Bayes estimator is further proposed to improve the precision of credit scoring by incorporating the dynamic interaction topology of clients. It is shown that under the proposed evolution framework, the designed estimator has a higher precision than any efficient estimator, and the mean square errors are strictly smaller than the Cram\'er-Rao lower bound for clients within a certain range of scores. Finally, simulation results for a special case illustrate the feasibility and effectiveness of the proposed algorithms.

Artificial Intelligence is one of the major tech innovations in commercial banking, aimed to automate and streamline ways banks are using, assessing risk and worthiness of credit decisions, identifying frauds among other applications. The methodology was structured in two phases: a quantitative analysis using institutional data, followed by a qualitative exploration through expert interviews. In the quantitative phase, secondary data were collected from five financial institutions, including three commercial banks and two fintech companies in Bangladesh. The qualitative phase was conducted to explore practical and regulatory challenges of AI implementation in real-world banking contexts. A comparative analysis between traditional and AI-based credit scoring systems revealed significant improvements across all key performance indicators. The AI-based system achieved a higher loan approval rate (78%) and a lower default rate (6%), while reducing processing time from 45 to 12 minutes and significantly enhancing customer satisfaction. Legacy system incompatibility was the most often mentioned barrier (67%), followed by problems with real-time data access (61%), and the inability of AI to explain itself (55%). However, when it comes to the usage of AI, banking professionals' top concerns are explainability (80%) and human oversight (88%). The results highlight the importance of using more advanced AI models, like XGBoost and Neural Networks, to improve the precision of credit evaluations and the efficiency of fraudulent transactions in real time, particularly for formerly underserved customer segments.

Under the present financial situation, accurate credit risk assessment is required for the sustainable development of lending institutions. According to the data of Lending Club, a well-known P2P lending website, this paper establishes a composite credit risk assessment model based on the combination of the analytic hierarchy process (AHP) and back propagation (BP) neural network. AHP is employed to rank and evaluate hierarchically the various factors affecting credit risk in a well-structured approach of risk factor analysis. The robust predictive power of the BP neural network is then used to establish a predictive model with the ability to identify patterns and correlations in the data. The model is constructed and validated on a large Lending Club dataset covering a wide range of borrower characteristics and loan performance outcomes. The study contributes to credit risk modeling by the systematic, data-driven approach, a novel credit scoring framework improving the accuracy and effectiveness of credit risk assessment and with the potential to enhance the lending decision-making process for institutions.

The banking industry faces significant challenges in balancing risk, speed, and precision when assessing creditworthiness. Traditional credit assessment methods often rely on rigid scoring systems that fail to adapt to dynamic market conditions, resulting in inefficiencies and heightened risks. This study focuses on developing an advanced machine learning (ML) decision-making model tailored for credit assessments in banking. By leveraging ML algorithms, the proposed model aims to enhance predictive accuracy, optimize decision-making speed, and mitigate risks associated with loan approvals and defaults. The research introduces a hybrid framework that integrates supervised learning techniques, such as gradient boosting and neural networks, with unsupervised learning methods for anomaly detection and clustering. These approaches enable the model to analyze large volumes of structured and unstructured data, including financial records, transaction histories, and behavioral patterns, to generate precise credit risk assessments. The model also incorporates explainable AI (XAI) techniques to ensure transparency and regulatory compliance, addressing a critical barrier to ML adoption in banking. Key findings highlight the model’s superior performance compared to traditional methods, achieving higher predictive accuracy, faster processing times, and improved risk management. Case studies from pilot implementations demonstrate its effectiveness in reducing non-performing loans, identifying high-risk borrowers, and enhancing customer experience through personalized credit offers. Furthermore, the research underscores the importance of robust data governance, algorithmic fairness, and cybersecurity in ensuring the reliability and ethical use of ML in banking. The proposed model provides a scalable and adaptable solution for banks to meet the evolving demands of modern financial ecosystems. By integrating real-time analytics and advanced decision-making capabilities, the model not only enhances operational efficiency but also supports long-term financial stability. This study contributes to the growing body of knowledge on artificial intelligence in financial services, offering actionable insights for financial institutions aiming to modernize their credit assessment processes.

Aiming at the problem of difficulty in identifying abnormal behaviors and low risk assessment accuracy in big data environments, a detection and assessment method framework based on graph neural network (GNN) is proposed. This method achieves deep expression of node features and abnormal identification of local patterns by constructing a behavior interaction graph and integrating a multi-layer graph neural structure. In the model design, a graph embedding reconstruction mechanism and a risk scoring function are introduced, and a complete process that can be applied to actual systems is constructed by combining the edge-cloud collaborative deployment architecture. Taking credit card fraud detection data as the experimental object, the model outperforms the isolation forest and traditional rule-based methods in indicators such as AUC and F1 Score, showing higher detection accuracy and stability. The results show that this method has good generalization ability and practical value in processing largescale structured and dynamic behavior data, and provides an efficient and feasible technical path for risk warning in the fields of financial risk control and network security.

APICURON is an open and freely accessible resource that tracks and credits the work of biocurators across multiple participating knowledgebases. Biocuration is essential to extract knowledge from research data and make it available in a structured and standardized way to the scientific community. However, processing biological data - mainly from literature - requires a huge effort which is difficult to quantify and acknowledge. APICURON collects biocuration events from third party resources and aggregates this information, spotlighting biocurator contributions. APICURON promotes biocurator engagement implementing gamification concepts like badges, medals and leaderboards and at the same time provides a monitoring service for registered resources and for biocurators themselves. APICURON adopts a data model that is flexible enough to represent and track the majority of biocuration activities. Biocurators are identified through their ORCID. The definition of curation events, scoring systems and rules for assigning badges and medals are resource-specific and easily customizable. Registered resources can transfer curation activities on the fly through a secure and robust Application Programming Interface (API). Here we show how simple and effective it is to connect a resource to APICURON describing the DisProt database of intrinsically disordered proteins as a use case. We believe APICURON will provide biological knowledgebases with a service to recognize and credit the effort of their biocurators, monitor their activity and promote curators engagement. Database URL: https://apicuron.org

We propose and study a minimalist approach towards synthetic tabular data generation. The model consists of a minimalistic unsupervised SparsePCA encoder (with contingent clustering step or log transformation to handle nonlinearity) and XGboost decoder which is SOTA for structured data regression and classification tasks. We study and contrast the methodologies with (variational) autoencoders in several toy low dimensional scenarios to derive necessary intuitions. The framework is applied to high dimensional simulated credit scoring data which parallels real-life financial applications. We applied the method to robustness testing to demonstrate practical use cases. The case study result suggests that the method provides an alternative to raw and quantile perturbation for model robustness testing. We show that the method is simplistic, guarantees interpretability all the way through, does not require extra tuning and provide unique benefits.

This study presents a comprehensive model for leveraging Artificial Intelligence (AI) and Big Data analytics to predict and mitigate financial risk in African markets, which are often characterized by volatility, data fragmentation, and limited transparency. The proposed model integrates machine learning algorithms with high-volume, high-velocity data streams sourced from diverse financial, economic, and socio-political datasets across the continent. By applying predictive analytics, the model identifies emerging risks and patterns in real time, enabling proactive decision-making by financial institutions, regulators, and investors. The framework is built around four core components: (1) data integration from structured and unstructured sources, including market transactions, news feeds, social media sentiment, and macroeconomic indicators; (2) machine learning models trained on historical data to forecast credit defaults, currency devaluation, inflation shocks, and systemic vulnerabilities; (3) a risk scoring engine that continuously updates probability metrics for various risk categories across sectors and countries; and (4) a user-friendly dashboard for visualization, scenario analysis, and strategic planning. The model was tested using data from five African economies Nigeria, Kenya, Ghana, South Africa, and Egypt covering a 10-year period. Results demonstrate its high predictive accuracy in detecting early warning signals for financial crises, currency instability, and stock market fluctuations. Moreover, the model’s capacity to process unstructured data, such as political discourse and policy changes, enhances its contextual intelligence in Africa’s dynamic financial environments. This research contributes to the development of localized, data-driven risk management systems in Africa, promoting financial inclusion, investment confidence, and regulatory innovation. It also addresses challenges such as data scarcity and reliability through hybrid approaches combining supervised learning, natural language processing, and human-in-the-loop methods. By equipping stakeholders with actionable insights, the model fosters a more resilient and transparent financial ecosystem in Africa.

Advancements in fintech algorithms have improved decision-making efficiency in credit scoring, investment advice, and financial product offerings. However, these automated systems raise ethical concerns related to algorithmic bias, lack of transparency, and accountability. Social inequalities embedded in historical data risk reinforcing discrimination in digital financial services, particularly in Southeast Asia’s evolving regulatory environment. This study explores ethical dilemmas in algorithmic decision-making across fintech platforms and assesses company responses. Using a qualitative multiple-case study of three Indonesian fintech firms in peer-to-peer lending, e-wallet, and robo-advisory sectors, data were gathered through semi-structured interviews and internal document analysis. Results indicate algorithmic bias as the most critical issue, followed by transparency and accountability gaps. Peer-to-peer lending firms demonstrate better ethical readiness via regular audits, while others show limited mitigation efforts. The study proposes a conceptual model emphasizing fairness, transparency, and accountability, offering practical insights for regulators and industry to strengthen ethical governance in Indonesia’s AI-based fintech ecosystem.

This study aims to examine and design a digital risk management model in ijarah financing at Bank Syariah Indonesia (BSI). The research is driven by the urgency of digital transformation in the Islamic banking sector while maintaining compliance with Sharia principles, particularly in risk management. The ijarah contract carries unique risk characteristics such as asset risk, moral hazard, and default risk, which must be systematically managed. This research adopts an exploratory qualitative approach through a case study at BSI, using semi-structured interviews and documentation as data collection methods. The findings indicate that the application of digital systems such as risk dashboards, credit scoring automation, and digital traceability significantly enhances risk mitigation in ijarah financing. The study proposes a Sharia-compliant digital risk management model that can be implemented adaptively and efficiently in BSI and other Islamic banks. The model not only strengthens risk control but also improves operational efficiency and transparency. This research contributes both practically and theoretically to the development of Sharia-based digital risk management practices.

Financial services firms encounter unparalleled expectations for agility, personalization, and compliance throughout the digital transformation age. Real-time data streaming, facilitated by technologies like Apache Kafka, Apache Spark Structured Streaming, Apache Flink, and Apache Pulsar, has become crucial for handling high-volume, high-velocity data streams. This article assesses these technologies, their architectural intricacies, performance attributes, and practical applications in trading, fraud detection, credit scoring, and compliance. The analysis utilizes industry documentation and scholarly sources to assist practitioners in implementing strategic real-time streaming solutions.

This study explores how Artificial Intelligence (AI) tools influence the operational efficiency of retail lending in commercial banks in Chennai. It focuses on four AI features: Automated Credit Scoring, AI-Powered Customer Support, Intelligent Loan Processing Systems, and Fraud Detection and Risk Analytics. Using a quantitative approach, data was collected from 200 digital-native customers who had availed retail loans, through a structured questionnaire and convenience sampling. Statistical analyses including regression and correlation were used to assess the impact of AI on operational efficiency. The results show that Automated Credit Scoring, AI Customer Support, and Intelligent Loan Processing significantly improve operational efficiency, while Fraud Detection had no notable impact. The findings help bridge the gap in regional, customer- centric research on AI in banking and provide valuable insights for banks aiming to improve service speed, accuracy, and customer satisfaction through AI-driven lending processes.

Credit risk evaluation is fundamental in financial decision-making, directly influencing lending strategies and default prevention. With the growing availability of structured financial data, machine learning methods have become increasingly prominent in building predictive credit scoring models. This research evaluates Decision Tree, Random Forest, and SVM classifiers for creditworthiness assessment. The Statlog (German Credit Data) dataset from UCI is used with a standardized preprocessing pipeline. Each model was trained and tested on the same dataset split and evaluated using standard classification metrics such as accuracy, precision, recall, and F1 score. Results show that the Random Forest classifier achieved the highest overall performance, particularly in identifying good credit applicants. At the same time, the Decision Tree maintained interpretability, and SVM offered a balanced trade-off. The findings highlight key considerations for model selection in credit scoring applications and suggest ensemble methods as strong candidates for future deployment.

Artificial Intelligence (AI) has been incorporated into the operations of Financial Technology (FINTECH) companies to streamline operations, render round-the-clock financial services and ensure efficiency. Notwithstanding the advantages of AI in the operations of FINTECH companies, there exist some perceived disadvantages. Therefore, this study examined the influence of AI on financial services delivery of FINTECH companies operating in Nigeria using primary data sourced from respondents through the use of structured questionnaire. Percentage and regression techniques were employed in the analysis of data. The results indicated that Robotic Process Automation, AI-powered Chatbots and Virtual Assistants, as well as AI Credit Scoring have significant positive influence on financial services delivery of FINTECH companies. Thus, it was concluded that the use of AI has enhanced financial services delivery of FINTECH companies in Nigeria. The study recommended that the application of Robotic Process Automation in FINTECH should be sustained and the conventional banks should also employ such technology to improve their financial services delivery. Also, AI-powered Chatbots and Virtual Assistants should be upgraded to handle complex customers’ complaints logically and accurately without frequent recourse to human customer agents. Finally, The AI Credit Scoring should be programmed to store and upgrade customers’ credit rating continuously. This would make it possible for old customers who complied duly in repayment of their previous loans, but have not taken another loan for a while, not to start from the lowest level of consideration whenever they are ready to request for another loan.

The increasing reliance on artificial intelligence (AI) in credit scoring has raised concerns about algorithmic bias and data privacy, necessitating robust cybersecurity risk assessment frameworks. This study investigates the role of cybersecurity risk assessment in mitigating these risks, utilizing multiple datasets, including the Home Mortgage Disclosure Act (HMDA) dataset, the Equifax Data Breach Report, the Financial Cybersecurity Incidents Database, and the MITRE ATT&CK Financial Sector Threat Intelligence Dataset. We employ statistical fairness metrics, Bayesian Probability Modeling, Markov Chain Analysis, and Monte Carlo Simulations to evaluate the extent of bias, privacy risks, and cybersecurity vulnerabilities. Findings reveal significant disparities in loan approvals, with Black applicants receiving approval rates 28% lower than White applicants (χ² = 59.83, p < 0.001), highlighting systemic bias in AI-driven credit scoring. Data privacy remains a pressing issue, as financial sector breaches affect an average of 5,069,760 individuals per incident. Insider threats pose the greatest risk, with a probability of 0.81 of leading to financial fraud. These findings underscore the urgency of integrating fairness-aware machine learning, enhancing regulatory compliance with AI governance policies, and deploying AI-driven cybersecurity tools to fortify financial AI applications against emerging threats. This research contributes to the broader discourse on ethical AI by providing a structured cybersecurity risk assessment approach to mitigate algorithmic bias and strengthen data privacy protections. Implementing these recommendations will enhance fairness, security, and transparency in AI-driven financial decision-making, ensuring compliance with evolving regulatory frameworks and fostering trust in automated credit scoring systems.

No abstract available

AutoML has revolutionized credit risk assessment in financial technology by automating feature engineering and pattern recognition. The integration of machine learning automation has transformed traditional risk assessment methodologies, enabling financial institutions to process vast amounts of data efficiently while improving accuracy in default prediction. Through advanced pattern recognition and automated feature selection, institutions can now identify subtle risk indicators and counter-intuitive payment behaviors that were previously undetectable. The implementation has resulted in substantial improvements in operational efficiency, cost reduction, and risk management effectiveness, while maintaining regulatory compliance and data security standards. Furthermore, the adoption of AutoML has enabled financial institutions to leverage sophisticated algorithms for real-time risk assessment, enhanced decision-making capabilities, and predictive modeling, leading to improved customer experiences and more precise credit evaluations across diverse market segments.

This research paper presents a comprehensive approach to build a data-driven credit risk model using machine learning techniques. Despite advancements in risk assessment methods, loan defaults remain a significant concern for financial institutions. This research work describes a novel and robust model architecture designed to tackle a credit risk modeling and scorecard prediction in the field. The research outlines a systematic approach and presents an innovative framework encompassing data cleaning, feature engineering, and evaluation using various matrices. The dataset is obtained from a peer-to-peer lending platform (Lending Club) having more than 450,000 features. Feature selection is conducted using Chi-squared test and ANOVA F-statistic. Subsequently, Weight of Evidence binning and feature engineering are detailed to optimize the predictive power of selected features. The model is trained using logistic regression with class weight balancing. Following this, a credit scorecard is developed based on logistic regression coefficients and Loan approval cut-offs are set to balance approval and rejection rates. The evaluation metrics, including AUROC, ROC, and PR curves are used to assess the model performance. The study concludes by highlighting the benefits of the credit risk model while acknowledging its limitations. The proposed model leverages state-of-the-art machine learning techniques and draws inspiration from various cutting-edge methodologies.

How to solve the accuracy problem in credit risk modeling and prediction in the field of financial technology, random forest algorithm, as a powerful machine learning method, has wide application potential in credit risk prediction. In this study, random forest algorithm is used as the main modeling and prediction tool. Through the collection of a large amount of credit-related data, the data is preprocessed and feature engineering, including missing value processing, feature selection, and data standardization. At the same time, a random forest model of multiple decision trees is constructed, and the model is trained and optimized. Evaluate the performance of the model and compare it. The accuracy rate of passing the test is as low as 90% and as high as 98%. This model can effectively identify potential default risks, so random forests can be applied in this field.

The advent of large language models has ushered in a new era of agentic systems, where artificial intelligence programs exhibit remarkable autonomous decision-making capabilities across diverse domains. This paper explores agentic system workflows in the financial services industry. In particular, we build agentic crews with human-in-the-loop module that can effectively collaborate to perform complex modeling and model risk management (MRM) tasks. The modeling crew consists of a judge agent and multiple agents who perform specific tasks such as exploratory data analysis, feature engineering, model selection/hyperparameter tuning, model training, model evaluation, and writing documentation. The MRM crew consists of a judge agent along with specialized agents who perform tasks such as checking compliance of modeling documentation, model replication, conceptual soundness, analysis of outcomes, and writing documentation. We demonstrate the effectiveness and robustness of modeling and MRM crews by presenting a series of numerical examples applied to credit card fraud detection, credit card approval, and portfolio credit risk modeling datasets.

With the development of the financial industry, credit default prediction, as an important task in financial risk management, has received increasing attention. Traditional credit default prediction methods mostly rely on machine learning models, such as decision trees and random forests, but these methods have certain limitations in processing complex data and capturing potential risk patterns. To this end, this paper proposes a deep learning model based on the combination of convolutional neural networks (CNN) and Transformer for credit user default prediction. The model combines the advantages of CNN in local feature extraction with the ability of Transformer in global dependency modeling, effectively improving the accuracy and robustness of credit default prediction. Through experiments on public credit default datasets, the results show that the CNN+Transformer model outperforms traditional machine learning models, such as random forests and XGBoost, in multiple evaluation indicators such as accuracy, AUC, and KS value, demonstrating its powerful ability in complex financial data modeling. Further experimental analysis shows that appropriate optimizer selection and learning rate adjustment play a vital role in improving model performance. In addition, the ablation experiment of the model verifies the advantages of the combination of CNN and Transformer and proves the complementarity of the two in credit default prediction. This study provides a new idea for credit default prediction and provides strong support for risk assessment and intelligent decision-making in the financial field. Future research can further improve the prediction effect and generalization ability by introducing more unstructured data and improving the model architecture.

In modern engineering, the importance of risk management in decision-making has increased proportionally with the growing scale and complexity of projects. Construction engineering projects are frequently faced with various risks, including insufficient financial support, human resource allocation difficulties, supply chain disruptions, etc., which may cause project delays, cost overruns, or even failure. Therefore, effective evaluation and mitigation of risks in different construction phases are crucial to ensuring project smooth progress. This study focuses on the challenge of intelligent risk assessment in construction engineering and proposes a novel evaluation model integrating deep neural network (DNN) modeling with intelligent decision calculus. The framework first uses the Analytic Hierarchy Process (AHP) to quantitatively measure phase-specific evaluation indicators, then performs feature extraction through Temporal Convolutional Networks (TCN). Reinforcement learning (via Deep Q-Networks, DQN) is incorporated to enhance the model’s interactive decision-making ability, realizing dynamic risk identification. Experimental results show that the model achieves an identification accuracy of over 80% in distinguishing low, medium, and high-risk scenarios, demonstrating an innovative approach to construction risk assessment that significantly improves decision-making efficiency.

No abstract available

Polycystic Ovary Syndrome (PCOS) is a common endocrine illness that affects women worldwide and can cause serious problems like infertility and an increased risk of miscarriage. It is often caused by high androgen and male hormone levels. This hormonal disorder disrupts the menstrual cycle, leading to irregular, delayed, or absent periods. PCOS can also contribute to severe health complications, including diabetes, gestational diabetes, weight gain, and excessive body hair. Artificial Intelligence (AI) has significantly transformed healthcare, enabling advancements in both science and engineering domains. To identify the most relevant features for PCOS detection Recursive Feature Elimination (RFE) is used, and several ensemble models such as bagging, boosting, voting classifiers, and stacking are compared. Among these, the Random Forest method using bagging achieved the highest accuracy. To enhance interpretability and assist medical professionals in decision-making, Explainable AI (XAI) techniques such as SHAP (SHapley Additive exPlanations) are applied to the Random Forest model that offers valuable insights into the factors influencing PCOS.

Abstract Regression is a fundamental prediction task common in data-centric engineering applications that involves learning mappings between continuous variables. In many engineering applications (e.g., structural health monitoring), feature-label pairs used to learn such mappings are of limited availability, which hinders the effectiveness of traditional supervised machine learning approaches. This paper proposes a methodology for overcoming the issue of data scarcity by combining active learning (AL) for regression with hierarchical Bayesian modeling. AL is an approach for preferentially acquiring feature-label pairs in a resource-efficient manner. In particular, the current work adopts a risk-informed approach that leverages contextual information associated with regression-based engineering decision-making tasks (e.g., inspection and maintenance). Hierarchical Bayesian modeling allow multiple related regression tasks to be learned over a population, capturing local and global effects. The information sharing facilitated by this modeling approach means that information acquired for one engineering system can improve predictive performance across the population. The proposed methodology is demonstrated using an experimental case study. Specifically, multiple regressions are performed over a population of machining tools, where the quantity of interest is the surface roughness of the workpieces. An inspection and maintenance decision process is defined using these regression tasks, which is in turn used to construct the active-learning algorithm. The novel methodology proposed is benchmarked against an uninformed approach to label acquisition and independent modeling of the regression tasks. It is shown that the proposed approach has superior performance in terms of expected cost—maintaining predictive performance while reducing the number of inspections required.

Abstract— Employee engagement and occupational burnout significantly influence workforce productivity and organizational sustainability. This research presents a structured analytical framework for evaluating employee engagement through a composite index approach and identifying burnout risk using rule-based classification logic. The study integrates data preprocessing, feature engineering, quantitative modeling, and interactive visualization to provide interpretable workforce intelligence. Experimental analysis demonstrates that overtime exposure and reduced work-life balance are strongly associated with elevated burnout risk. The proposed framework offers a scalable and transparent decision-support model suitable for modern HR analytics environments. Keywords— Employee Engagement, Burnout Risk, HR Analytics, Composite Index, Workforce Intelligence, Data Visualization.

: Heart disease prediction is a critical issue in healthcare, where accurate early diagnosis can save lives and reduce healthcare costs. The problem is inherently complex due to the high dimensionality of medical data, irrelevant or redundant features, and the variability in risk factors such as age, lifestyle, and medical history. These challenges often lead to inefficient and less accurate models. Traditional prediction methodologies face limitations in effectively handling large feature sets and optimizing classification performance, which can result in overfitting poor generalization, and high computational cost. This work proposes a novel classification model for heart disease prediction that addresses these challenges by integrating feature selection through a Genetic Algorithm (GA) with an ensemble deep learning approach optimized using the Tunicate Swarm Algorithm (TSA). GA selects the most relevant features, reducing dimensionality and improving model efficiency. The selected features are then used to train an ensemble of deep learning models, where the TSA optimizes the weight of each model in the ensemble to enhance prediction accuracy. This hybrid approach addresses key challenges in the field, such as high dimensionality, redundant features, and classification performance, by introducing an efficient feature selection mechanism and optimizing the weighting of deep learning models in the ensemble. These enhancements result in a model that achieves superior accuracy, generalization, and efficiency compared to traditional methods. The proposed model demonstrated notable advancements in both prediction accuracy and computational efficiency over traditional models. Specifically, it achieved an accuracy of 97.5%, a sensitivity of 97.2%, and a specificity of 97.8%. Additionally, with a 60–40 data split and 5-fold cross-validation, the model showed a significant reduction in training time (90 s), memory consumption (950 MB), and CPU usage (80%), highlighting its effectiveness in processing large, complex medical datasets for heart disease prediction.

A growing elderly population suffering from incurable, chronic conditions such as dementia present a continual strain on medical services due to mental impairment paired with high comorbidity resulting in increased hospitalization risk. The identification of at risk individuals allows for preventative measures to alleviate said strain. Electronic health records provide opportunity for big data analysis to address such applications. Such data however, provides a challenging problem space for traditional statistics and machine learning due to high dimensionality and sparse data elements. This article proposes a novel machine learning methodology: entropy regularization with ensemble deep neural networks (ECNN), which simultaneously provides high predictive performance of hospitalization of patients with dementia whilst enabling an interpretable heuristic analysis of the model architecture, able to identify individual features of importance within a large feature domain space. Experimental results on health records containing 54,647 features were able to identify 10 event indicators within a patient timeline: a collection of diagnostic events, medication prescriptions and procedural events, the highest ranked being essential hypertension. The resulting subset was still able to provide a highly competitive hospitalization prediction (Accuracy: 0.759) as compared to the full feature domain (Accuracy: 0.755) or traditional feature selection techniques (Accuracy: 0.737), a significant reduction in feature size. The discovery and heuristic evidence of correlation provide evidence for further clinical study of said medical events as potential novel indicators. There also remains great potential for adaption of ECNN within other medical big data domains as a data mining tool for novel risk factor identification.

To address the challenges in forecasting crude oil and hot-rolled coil futures prices, the aim is to transcend the constraints of conventional approaches. This involves effectively predicting short-term price fluctuations, developing quantitative trading strategies, and modeling time series data. The goal is to enhance prediction accuracy and stability, thereby supporting decision-making and risk management in financial markets. A novel approach, the multi-dimensional fusion feature-enhanced (MDFFE) prediction method has been devised. Additionally, a data augmentation framework leveraging multi-dimensional feature engineering has been established. The technical indicators, volatility indicators, time features, and cross-variety linkage features are integrated to build a prediction system, and the lag feature design is used to prevent data leakage. In addition, a deep fusion model is constructed, which combines the temporal feature extraction ability of the convolution neural network with the nonlinear mapping advantage of an extreme gradient boosting tree. With the help of a three-layer convolution neural network structure and adaptive weight fusion strategy, an end-to-end prediction framework is constructed. Experimental results demonstrate that the MDFFE model excels in various metrics, including mean absolute error, root mean square error, mean absolute percentage error, coefficient of determination, and sum of squared errors. The mean absolute error reaches as low as 0.0068, while the coefficient of determination can be as high as 0.9970. In addition, the significance and stability of the model performance were verified by statistical methods such as a paired t-test and ANOVA analysis of variance. This MDFFE algorithm offers a robust and practical approach for predicting commodity futures prices. It holds significant theoretical and practical value in financial market forecasting, enhancing prediction accuracy and mitigating forecast volatility.

Introduction Hospital-acquired infections of communicable viral diseases (CVDs) have been posing a tremendous challenge to healthcare workers globally. Healthcare personnel (HCP) is facing a consistent risk of viral infections, and subsequently higher rates of morbidity and mortality. Materials and methods We proposed a domain-knowledge-driven infection risk model to quantify the individual HCP and the population-level risks. For individual-level risk estimation, a time-variant infection risk model is proposed to capture the transmission dynamics of CVDs. At the population-level, the infection risk is estimated using a Bayesian network model constructed from three feature sets, including individual-level factors, engineering control factors, and administrative control factors. For model validation, we investigated the case study of the Coronavirus disease, in which the individual-level and population-level infection risk models were applied. The data were collected from various sources such as COVID-19 transmission databases, health surveys/questionaries from medical centers, U.S. Department of Labor databases, and cross-sectional studies. Results Regarding the individual-level risk model, the variance-based sensitivity analysis indicated that the uncertainty in the estimated risk was attributed to two variables: the number of close contacts and the viral transmission probability. Next, the disease transmission probability was computed using a multivariate logistic regression applied for a cross-sectional HCP data in the UK, with the 10-fold cross-validation accuracy of 78.23%. Combined with the previous result, we further validated the individual infection risk model by considering six occupations in the U.S. Department of Labor O*Net database. The occupation-specific risk evaluation suggested that the registered nurses, medical assistants, and respiratory therapists were the highest-risk occupations. For the population-level risk model validation, the infection risk in Texas and California was estimated, in which the infection risk in Texas was lower than that in California. This can be explained by California’s higher patient load for each HCP per day and lower personal protective equipment (PPE) sufficiency level. Conclusion The accurate estimation of infection risk at both individual level and population levels using our domain-knowledge-driven infection risk model will significantly enhance the PPE allocation, safety plans for HCP, and hospital staffing strategies.

Machine learning (ML) offers transformative potential in public health by enabling predictive modeling of complex disease interactions, including co-infections. Tuberculosis (TB) remains a major global health challenge, with co-infection among high-risk populations—particularly HIV-positive individuals—significantly increasing morbidity and mortality. Early identification of individuals at heightened risk of TB co-infection is critical for timely interventions, targeted screening, and optimized clinical management. This presents a machine learning framework for predicting TB co-infection risk among high-risk populations, leveraging demographic, clinical, immunological, and behavioral data to enhance predictive accuracy. The model incorporates key predictors, including HIV viral load, CD4 counts, antiretroviral therapy adherence, socioeconomic factors, comorbidities, and exposure-related variables. Feature engineering and selection methods are employed to identify the most informative variables, while class imbalance techniques address the relative rarity of TB co-infection events. Supervised machine learning algorithms, including random forests, gradient boosting, and logistic regression ensembles, are applied to training datasets drawn from national health registries, cohort studies, and electronic medical records. Model performance is evaluated through cross-validation, with metrics including area under the receiver operating characteristic curve (AUC), precision, recall, and F1 score to ensure robustness and generalizability. Preliminary findings indicate that integrating clinical and sociodemographic data substantially improves risk stratification, enabling healthcare providers to prioritize screening and preventive therapy for high-risk individuals. Additionally, the model demonstrates the potential to uncover non-linear relationships and interactions between immunological and behavioral factors that may be overlooked in traditional statistical analyses. Beyond predictive accuracy, the framework emphasizes interpretability, transparency, and ethical considerations, ensuring that predictions support clinical decision-making without reinforcing biases or inequities in care. Machine learning models offer a promising approach for anticipating TB co-infection risk among vulnerable populations, particularly HIV-positive individuals. By combining high-dimensional data with advanced predictive algorithms, these models can inform proactive public health strategies, optimize resource allocation, and ultimately reduce TB-related morbidity and mortality. The framework underscores the importance of integrating data-driven tools into targeted interventions and surveillance systems for high-risk groups. Keywords: Tuberculosis, HIV, TB/HIV Co-Infection, High-Risk Populations, Predictive Modeling, Machine Learning, Risk Factors, Clinical Data, Demographic Data, Epidemiology.

The advent of large language models has ushered in a new era of agentic systems, where artificial intelligence programs exhibit remarkable autonomous decision-making capabilities across diverse domains. This paper explores agentic system workflows in the financial services industry. In particular, we build agentic crews with human-in-the-loop orchestrator that can effectively collaborate to perform complex machine learning modeling tasks. The modeling crew consists of a judge agent and multiple agents who perform specific tasks such as exploratory data analysis, feature engineering, model selection/hyperparameter tuning, model training, model evaluation, and writing documentation. We demonstrate the effectiveness and robustness of modeling crews by presenting a comparative experiment applied to the detection of credit card fraud and card portfolio credit risk. Our fraud detection experiment achieved a recall of 81.6 % and an F1-score of 88.9 %, outperforming AutoML's 70.4 % and 82.1 %, respectively.

The integration of machine learning (ML) into financial modeling represents a transformative shift in the finance sector, enhancing analytical capabilities and decision-making processes. This article explores the fundamental concepts of financial modeling and machine learning, illustrating how ML methodologies—both supervised and unsupervised—improve traditional financial models by enabling the analysis of extensive datasets to uncover intricate patterns. By leveraging historical data, ML enhances predictive accuracy in various applications, including risk assessment, algorithmic trading, credit scoring, and fraud detection. The article highlights significant advancements in predictive analytics through techniques such as regression analysis and time series forecasting, which allow financial analysts to navigate market uncertainties more effectively. Moreover, the automation of data processing and feature engineering through ML leads to increased operational efficiency, reducing human error and enhancing real-time decision-making capabilities. However, challenges such as data quality issues, model interpretability, and ethical considerations regarding algorithmic bias necessitate careful management to ensure responsible implementation. The article emphasizes the importance of transparency in ML models to foster trust among stakeholders while meeting regulatory requirements. Future research directions are proposed, focusing on the integration of explainable AI with blockchain technology and the implications of big data analytics on ML models. By addressing these emerging challenges and opportunities, finance professionals can leverage advanced analytical tools to drive innovation and enhance strategic initiatives in an increasingly complex financial landscape. Ultimately, this article underscores the pivotal role of machine learning in reshaping financial modeling practices and its potential to redefine the future of finance.

Customers are prominent resources in every business for its sustainability. Therefore, predicting customer churn is significant for reducing churn, particularly in the high-churn-rate telecommunications business. To identify customers at risk of churning, tactical marketing actions can be strategized to raise the likelihood of the churn-probable customers remaining as customers. This might provide a corporation with significant savings. Hence, in this work, a churn prediction system is developed to assist telecommunication operators in detecting potential churn customers. In the proposed framework, the input data quality is improved through the processes of exploratory data analysis and data preprocessing for identifying data errors and comprehending data patterns. Then, feature engineering and data sampling processes are performed to transform the captured data into an appropriate form for classification and imbalanced data handling. An optimized ensemble learning model is proposed for classification in this framework. Unlike other ensemble models, the proposed classification model is an optimized weighted soft voting ensemble with a sequence of weights applied to weigh the prediction of each base learner with the hypothesis that specific base learners in the ensemble have more skill than others. In this optimization, Powell’s optimization algorithm is applied to optimize the ensemble weights of influence according to the base learners’ importance. The efficiency of the proposed optimally weighted ensemble learning model is evaluated in a real-world database. The empirical results show that the proposed customer churn prediction system achieves a promising performance with an accuracy score of 84% and an F1 score of 83.42%. Existing customer churn prediction systems are studied. We achieved a higher prediction accuracy than the other systems, including machine learning and deep learning models.

Deep neural networks (DNNs) have demonstrated superior performance over classical machine learning to support many features in safety-critical systems. Although DNNs are now widely used in such systems (e.g., self driving cars), there is limited progress regarding automated support for functional safety analysis in DNN-based systems. For example, the identification of root causes of errors, to enable both risk analysis and DNN retraining, remains an open problem. In this article, we propose SAFE, a black-box approach to automatically characterize the root causes of DNN errors. SAFE relies on a transfer learning model pre-trained on ImageNet to extract the features from error-inducing images. It then applies a density-based clustering algorithm to detect arbitrary shaped clusters of images modeling plausible causes of error. Last, clusters are used to effectively retrain and improve the DNN. The black-box nature of SAFE is motivated by our objective not to require changes or even access to the DNN internals to facilitate adoption. Experimental results show the superior ability of SAFE in identifying different root causes of DNN errors based on case studies in the automotive domain. It also yields significant improvements in DNN accuracy after retraining, while saving significant execution time and memory when compared to alternatives.

No abstract available

Feature engineering is a time consuming component of predictive modeling. We propose a versatile platform to automatically extract features for risk prediction, based on a pre-defined and extensible entity schema. The extraction is independent of disease type or risk prediction task. We contrast auto-extracted features to baselines generated from the Elixhauser comorbidities. Hospital medical records was transformed to event sequences, to which filters were applied to extract feature sets capturing diversity in temporal scales and data types. The features were evaluated on a readmission prediction task, comparing with baseline feature sets generated from the Elixhauser comorbidities. The prediction model was through logistic regression with elastic net regularization. Predictions horizons of 1, 2, 3, 6, 12 months were considered for four diverse diseases: diabetes, COPD, mental disorders and pneumonia, with derivation and validation cohorts defined on non-overlapping data-collection periods. For unplanned readmissions, auto-extracted feature set using socio-demographic information and medical records, outperformed baselines derived from the socio-demographic information and Elixhauser comorbidities, over 20 settings (5 prediction horizons over 4 diseases). In particular over 30-day prediction, the AUCs are: COPD—baseline: 0.60 (95% CI: 0.57, 0.63), auto-extracted: 0.67 (0.64, 0.70); diabetes—baseline: 0.60 (0.58, 0.63), auto-extracted: 0.67 (0.64, 0.69); mental disorders—baseline: 0.57 (0.54, 0.60), auto-extracted: 0.69 (0.64,0.70); pneumonia—baseline: 0.61 (0.59, 0.63), auto-extracted: 0.70 (0.67, 0.72). The advantages of auto-extracted standard features from complex medical records, in a disease and task agnostic manner were demonstrated. Auto-extracted features have good predictive power over multiple time horizons. Such feature sets have potential to form the foundation of complex automated analytic tasks.

Landmines remain a threat to war-affected communities for years after conflicts have ended, partly due to the laborious nature of demining tasks. Humanitarian demining operations begin by collecting relevant information from the sites to be cleared, which is then analyzed by human experts to determine the potential risk of remaining landmines. In this article, we propose RELand system to support these tasks, which consists of three major components. We (1) provide general feature engineering and label assigning guidelines to enhance datasets for landmine risk modeling, which are widely applicable to global demining routines, (2) formulate landmine presence as a classification problem and design a novel interpretable model based on sparse feature masking and invariant risk minimization, and run extensive evaluation under proper protocols that resemble real-world demining operations to show a significant improvement over the state-of-the-art, and (3) build an interactive web interface to suggest priority areas for demining organizations. We are currently collaborating with a humanitarian demining NGO in Colombia that is using our system as part of their field operations in two areas recently prioritized for demining. The resulting dataset and developed code can be found here.

In many real-world datasets, rows may have distinct characteristics and require different modeling approaches for accurate predictions. In this paper, we propose an adaptive modeling approach for row-type dependent predictive analysis(RTDPA). Our framework enables the development of models that can effectively handle diverse row types within a single dataset. Our dataset from XXX bank contains two different risk categories, personal loan and agriculture loan. each of them are categorised into four classes standard, sub-standard, doubtful and loss. We performed tailored data pre processing and feature engineering to different row types. We selected traditional machine learning predictive models and advanced ensemble techniques. Our findings indicate that all predictive approaches consistently achieve a precision rate of no less than 90%. For RTDPA, the algorithms are applied separately for each row type, allowing the models to capture the specific patterns and characteristics of each row type. This approach enables targeted predictions based on the row type, providing a more accurate and tailored classification for the given dataset.Additionally, the suggested model consistently offers decision makers valuable and enduring insights that are strategic in nature in banking sector.

The decision-making process in real-world implementations has been affected by a growing reliance on data-driven models. We investigated the synergetic pattern between the data-driven methods, empirical domain knowledge, and first-principles simulations. We showed the potential risk of biased results when using data-driven models without causal analysis. Using a case study assessing the implication of several design solutions on the energy consumption of a building, we proved the necessity of causal analysis during the data-driven modeling process. We concluded that: (a) Data-driven models' accuracy assessment or domain knowledge screening may not rule out biased and spurious results; (b) Data-driven models' feature selection should involve careful consideration of causal relationships, especially colliders; (c) Causal analysis results can be used as an aid to first-principles simulation design and parameter checking to avoid cognitive biases. We proved the benefits of causal analysis when applied to data-driven models in building engineering.

For reliability, machine learning models in some areas, e.g., finance and healthcare, require to be both accurate and globally interpretable. Among them, credit risk assessment is a major application of machine learning for financial institutions to evaluate credit of users and detect default or fraud. Simple white-box models, such as Logistic Regression (LR), are usually used for credit risk assessment, but not powerful enough to model complex nonlinear interactions among features. In contrast, complex black-box models are powerful at modeling, but lack of interpretability, especially global interpretability. Fortunately, automatic feature crossing is a promising way to find cross features to make simple classifiers to be more accurate without heavy handcrafted feature engineering. However, existing automatic feature crossing methods have problems in efficiency on credit risk assessment, for corresponding data usually contains hundreds of feature fields. In this work, we find local interpretations in Deep Neural Networks (DNNs) of a specific feature are usually inconsistent among different samples. We demonstrate this is caused by nonlinear feature interactions in the hidden layers of DNN. Thus, we can mine feature interactions in DNN, and use them as cross features in LR. This will result in mining cross features more efficiently. Accordingly, we propose a novel automatic feature crossing method called DNN2LR. The final model, which is a LR model empowered with cross features, generated by DNN2LR is a white-box model. We conduct experiments on both public and business datasets from real-world credit risk assessment applications, which show that, DNN2LR outperform both conventional models used for credit assessment and several feature crossing methods. Moreover, comparing with state-of-the-art feature crossing methods, i.e., AutoCross, the proposed DNN2LR method accelerates the speed by about 10 to 40 times on financial credit assessment datasets, which contain hundreds of feature fields.

Graph Convolutional Network (GCN) is an emerging technique that performs learning and reasoning on graph data. It operates feature learning on the graph structure, through aggregating the features of the neighbor nodes to obtain the embedding of each target node. Owing to the strong representation power, recent research shows that GCN achieves state-of-the-art performance on several tasks such as recommendation and linked document classification. Despite its effectiveness, we argue that existing designs of GCN forgo modeling cross features, making GCN less effective for tasks or data where cross features are important. Although neural network can approximate any continuous function, including the multiplication operator for modeling feature crosses, it can be rather inefficient to do so (i.e., wasting many parameters at the risk of overfitting) if there is no explicit design. To this end, we design a new operator named Cross-feature Graph Convolution, which explicitly models the arbitrary-order cross features with complexity linear to feature dimension and order size. We term our proposed architecture as Cross-GCN, and conduct experiments on three graphs to validate its effectiveness. Extensive analysis validates the utility of explicitly modeling cross features in GCN, especially for feature learning at lower layers.

This article provides an analysis of the utilization of Machine Learning (ML) models in Flood Susceptibility Mapping (FSM), based on selected publications from the past decade (2013–2023). Recognizing the challenge that some stages of ML modeling inherently rely on experience or trial‐and‐error approaches, this work aims at establishing a clear roadmap for the deployment of ML‐based FSM frameworks. The critical aspects of ML‐based FSM are identified, including data considerations, the model's development procedure, and employed algorithms. A comparative analysis of different ML models, alongside their practical applications, is made. Findings suggest that despite existing limitations, ML methods, when carefully designed and implemented, can be successfully utilized to determine areas at risk of flooding. We show that the effectiveness of ML‐based FSM models is significantly influenced by data preprocessing, feature engineering, and the development of the model using the most impactful parameters, as well as the selection of the appropriate model type and configuration. Additionally, we introduce a structured roadmap for ML‐based FSM, identification of overlooked conditioning factors, comparative model analysis, and integration of practical considerations, all aimed at enhancing modeling quality and effectiveness. This comprehensive analysis thereby serves as a critical resource for professionals in the field of FSM.

No abstract available

Objectives To describe the processes developed by The Hospital for Sick Children (SickKids) to enable utilization of electronic health record (EHR) data by creating sequentially transformed schemas for use across multiple user types. Methods We used Microsoft Azure as the cloud service provider and named this effort the SickKids Enterprise-wide Data in Azure Repository (SEDAR). Epic Clarity data from on-premises was copied to a virtual network in Microsoft Azure. Three sequential schemas were developed. The Filtered Schema added a filter to retain only SickKids and valid patients. The Curated Schema created a data structure that was easier to navigate and query. Each table contained a logical unit such as patients, hospital encounters or laboratory tests. Data validation of randomly sampled observations in the Curated Schema was performed. The SK-OMOP Schema was designed to facilitate research and machine learning. Two individuals mapped medical elements to standard Observational Medical Outcomes Partnership (OMOP) concepts. Results A copy of Clarity data was transferred to Microsoft Azure and updated each night using log shipping. The Filtered Schema and Curated Schema were implemented as stored procedures and executed each night with incremental updates or full loads. Data validation required up to 16 iterations for each Curated Schema table. OMOP concept mapping achieved at least 80 % coverage for each SK-OMOP table. Conclusions We described our experience in creating three sequential schemas to address different EHR data access requirements. Future work should consider replicating this approach at other institutions to determine whether approaches are generalizable.

Tabular data pervades the landscape of the World Wide Web, playing a foundational role in the digital architecture that underpins online information. Given the recent influence of large-scale pretrained models like ChatGPT and SAM across various domains, exploring the application of pretraining techniques for mining tabular data on the web has emerged as a highly promising research direction. Indeed, there have been some recent works around this topic where most (if not all) of them are limited in the scope of a fixed-schema/single table. Due to the scale of the dataset and the parameter size of the prior models, we believe that we have not reached the ''BERT moment'' for the ubiquitous tabular data. The development on this line significantly lags behind the counterpart research domains such as natural language processing. In this work, we first identify the crucial challenges behind tabular data pretraining, particularly overcoming the cross-table hurdle. As a pioneering endeavor, this work mainly (i)-contributes a high-quality real-world tabular dataset, (ii)-proposes an innovative, generic, and efficient cross-table pretraining framework, dubbed as CM2, where the core to it comprises a semantic-aware tabular neural network that uniformly encodes heterogeneous tables without much restriction and (iii)-introduces a novel pretraining objective --- prompt Masked Table Modeling (pMTM) --- inspired by NLP but intricately tailored to scalable pretraining on tables. Our extensive experiments demonstrate CM2's state-of-the-art performance and validate that cross-table pretraining can enhance various downstream tasks.

The article examines the international experience of developing digital innovations in the credit market. It is established that digital innovations cover a wide range of technologies and their applications that transform the credit market in several key areas, namely: advanced technologies for assessing risks and solutions that are used to create alternative credit scoring models, security and optimization of operations can be used to ensure transparency and immutability of credit transaction records, reduce operating costs and increase data security, improve customer experience and accessibility increases the accessibility of financial services to a wide range of people. The classification of digital innovations is graphically presented, which includes the following stages: technological approach, represented by artificial intelligence, Blockchain, BigData, Mobile applications and FinTech, Cloud technologies; by the level of user experience, it is represented by components: innovations that require a high level of technological awareness and innovations that can be used by users with different levels of awareness; by level of regulation, it is represented by components: free-to-use innovations, innovations that require appropriate permits from regulatory authorities; by level of complexity, it is represented by components: easy-to-use and difficult-to-use and require significant investments. The table shows a statistical sample of the dynamics of the share of the EU population that used Internet-Banking during 2014-2024. The global territorial distribution of cloud computing implementation on the global market and the total volume of the global cloud computing market are graphically presented. The impact of fintech solutions, artificial intelligence, blockchain technologies and mobile applications on the transformation of credit markets is analyzed. The main trends, successful practices and potential risks associated with the implementation of digital innovations in lending are identified. Particular attention is paid to the analysis of regulatory barriers, security and privacy issues, insufficient financial literacy and lack of standardization, which hinder the spread of digital technologies in the credit sector.  

This article presents the research and development of a system for managing the effectiveness of organizational processes through the application of the concept of a Balanced Scorecard (BSC). The main goal is to create an integrated software system that combines the principles of BSC with the functionalities of modern ERP and Business Intelligence (BI) solutions. The project offers a three-layer architecture based on data from Microsoft Dynamics NAV/LS Central, analyzed and visualized using the BI tool Tableau. The central place is occupied by the database, which provides storage, unification and processing of key performance indicators (KPI). An ana-lytical model is applied to it, which allows multidimensional analysis by various criteria - departments, periods, employees and goals. By using structured database analysis, traceability of changes is ensured, the ability to predict and detect dependencies between different indicators. This allows management to make informed and justified decisions in real time. The developed concept is based on a systematic and analytical approach, includ-ing database modeling, defining relationships between key tables (Employees, Indicators, Results, Targets) and integration with ERP sources. The system provides a toolkit for automatic data extraction, processing and visu- alization through a BI platform. The expected contribution of the project is related to improving management accountability, increasing the transparency of organizational processes and creating a methodology for data analysis applicable in various business environments. The developed system can be used as a basis for a disser-tation aimed at implementing an intelligent platform for performance management through database analysis and integration with ERP systems. DOI

Credit risk modeling relies extensively on Weight of Evidence (WoE) and Information Value (IV) for feature engineering, and Population Stability Index (PSI) for drift monitoring, yet their theoretical foundations remain disconnected. We establish a unified information-theoretic framework revealing these industry-standard metrics as instances of classical information divergences. Specifically, we prove that IV exactly equals PSI (Jeffreys divergence) computed between good and bad credit outcomes over identical bins. Through the delta method applied to WoE transformations, we derive standard errors for IV and PSI, enabling formal hypothesis testing and probabilistic fairness constraints for the first time. We formalize credit modeling's inherent performance-fairness trade-off as maximizing IV for predictive power while minimizing IV for protected attributes. Using automated binning with depth-1 XGBoost stumps, we compare three encoding strategies: logistic regression with one-hot encoding, WoE transformation, and constrained XGBoost. All methods achieve comparable predictive performance (AUC 0.82-0.84), demonstrating that principled, information-theoretic binning outweighs encoding choice. Mixed-integer programming traces Pareto-efficient solutions along the performance-fairness frontier with uncertainty quantification. This framework bridges theory and practice, providing the first rigorous statistical foundation for widely-used credit risk metrics while offering principled tools for balancing accuracy and fairness in regulated environments.

Credit risk assessment is essential for financial institutions to effectively manage loan portfolios, especially for installment loans. Predicting delinquency is challenging due to the complex interplay of borrower behavior, loan characteristics, and repayment pattern. Traditional models often fail to capture non-linear relationships in data and require significant preprocessing to address imbalanced datasets, feature scaling, and diverse data distributions, resulting in inefficiencies. This research predicts installment loan delinquency using LightGBM, a gradient-boosting algorithm tailored for complex, imbalanced financial datasets. Unlike previous studies focusing on general credit risk or credit card defaults, this work specifically addresses the temporal and behavioral dynamics of installment loans. The model uses a real-world dataset from financial institutions, integrating borrower demographics, loan attributes, and engineered repayment features. LightGBM's histogram-based binning and inherent handling of heterogeneous feature scales both reduce preprocessing complexity and improve performance. Evaluation results show significant improvements over traditional models, achieving an AUC-ROC of 0.91 and strong precision and recall. This approach demonstrates scalability and effectiveness for modern credit risk management. Future work could incorporate macroeconomic factors and assess real-time deployment to further expand the model’s applicability.

Financial institutions use various data mining algorithms to determine the credit limits for individuals using features like age, education, employment, gender, income, and marital status. But, there is still a question of accurate predictability, that is, how accurate can an institution be in predicting risk and granting credit levels. If an institution grants too low of a credit limit/loan for an individual, then the institution may lose business to competitors, but if the institution grants too high of a credit limit/loan, then the institution may lose money if that individual does not repay the credit/loan. The novelty of this work is that it shows how to improve the accuracy in predicting credit limits/loan amounts using synthetic feature generation. By creating secondary groupings and including both the original binning and the synthetic bins, the classification accuracy and other statistical measures like precision and ROC improved substantially. Hence, our research showed that without synthetic feature generation, the classification rates were low, and the use of synthetic features greatly improved the classification accuracy and other statistical measures.

Loans are the primary revenue generator for banks because they earn interest income from the credit, they extend through lending products. However, defaults on these loans can significantly impact profits. By identifying borrowers likely to default, banks can mitigate risk and reduce non- performing loans in their portfolio. This makes the study of this phenomenon very important. Previous research has shown there are many methods to study loan default prediction, which is essential for maximizing profits. However, comparing the nature and performance of different techniques is critical for reliability. The project focuses on leveraging machine learning techniques to enhance the efficiency and accuracy of loan approval processes in financial institutions. By analyzing a dataset comprising various applicant attributes and historical loan data, predictive models are developed to assess the likelihood of loan repayment or default. This project will stand out by using multiple feature engineering techniques such as Binning and Bucketing, Polynomial, Interaction Features to enhance the dataset. Through multiple feature engineering, model evaluations and ensembling techniques, the project aims to provide a comprehensive solution for automating and optimizing loan approval decisions. Keywords: Loans approval, Machine Learning.

Credit risk prediction is a crucial task for financial institutions. The technological advancements in machine learning, coupled with the availability of data and computing power, has given rise to more credit risk prediction models in financial institutions. In this paper, we propose a stacked classifier approach coupled with a filter-based feature selection (FS) technique to achieve efficient credit risk prediction using multiple datasets. The proposed stacked model includes the following base estimators: Random Forest (RF), Gradient Boosting (GB), and Extreme Gradient Boosting (XGB). Furthermore, the estimators in the Stacked architecture were linked sequentially to extract the best performance. The filter- based FS method that is used in this research is based on information gain (IG) theory. The proposed algorithm was evaluated using the accuracy, the F1-Score and the Area Under the Curve (AUC). Furthermore, the Stacked algorithm was compared to the following methods: Artificial Neural Network (ANN), Decision Tree (DT), and k-Nearest Neighbour (KNN). The experimental results show that stacked model obtained AUCs of 0.934, 0.944 and 0.870 on the Australian, German and Taiwan datasets, respectively. These results, in conjunction with the accuracy and F1-score metrics, demonstrated that the proposed stacked classifier outperforms the individual estimators and other existing methods.

To solve the high-dimensional issue in credit risk assessment, a hybrid clustering and boosting tree feature selection method is proposed. In the hybrid methodology, an improved minimum spanning tree model is first used to remove redundant and irrelevant  features. Then three embedded feature selection approaches (i.e., Random Forest, XGBoost, and AdaBoost) are used to further enhance the feature-ranking efficiency and obtain better  prediction performance by applying the optimal features. For verification purpose, two real-world credit datasets are used to demonstrate the effectiveness of the proposed hybrid clustering and boosting tree feature selection (CBTFS) methodology. Experimental results demonstrated that the proposed method is superior to others classic feature selection methods. This indicates that the proposed hybrid clustering and boosting tree feature selection method can be used as a promising tool for solving high-dimensional issue in credit risk assessment. First published online 12 February 2025

No abstract available

Financial institutions increasingly rely on machine learning (ML) models to assess credit risk and make lending decisions. Accurate prediction hinges on effective feature selection, which can significantly enhance model performance. This paper investigates the efficacy of seven supervised ML algorithms in predicting credit risk: Naive Bayes, Support Vector Machine, Decision Tree, K-Nearest Neighbor, Artificial Neural Network, Random Forest, and Logistic Regression. Using a German credit dataset comprising 1000 observations with 20 explanatory variables, we evaluated model performance using accuracy, kappa statistic, and F1 score. Two data-splitting scenarios (70-30\% and 80-20\%) were employed to assess robustness. To optimize model performance, we addressed outliers through imputation methods and applied the Boruta algorithm for feature selection, which identified and eliminated six non-contributing features. Our findings consistently demonstrate the superiority of the Random Forest algorithm across both scenarios. In terms of accuracy, Random Forest achieved 77.3\% in the 70-30\% split and 80\% in the 80-20\% split, outperforming all other methods. These results underscore the potential of Random Forest as a valuable tool for credit risk assessment in financial institutions.

In the rapidly evolving digital finance landscape, accurate credit risk assessment is essential to mitigate financial losses and improve lending decisions. Traditional credit risk models often struggle with high-dimensional data and complex borrower behavior, leading to suboptimal predictions. This study addresses these challenges by an AI-driven credit risk assessment model is proposed that integrates advanced feature optimization with reinforcement learning. The DenseNet architecture is used for comprehensive feature analysis, while the enhanced orca predation (EOP) algorithm dynamically optimizes and filters key borrower attributes to highlight critical financial risk factors. Deep Q-learning model is used for effective credit risk evaluation, capturing non-linear patterns in borrower data. The proposed model is validated using the LendingClub online loan dataset, achieving an RMSE of 10.350, MSE of 107.000, R2 of 0.997, and RGA of 0.965, outperforming five state-of-the-art baseline models. Comparative analysis indicates that the proposed model reduces RMSE by 10.23 % and MSE by 19.55 % while improving R2 by 0.90 % and RGA by 1.27 % compared to the HBA-LGBM model. These results demonstrate that the proposed model effectively prioritizes critical borrower risk factors, reduces prediction errors, and provides a robust, scalable, and interpretable solution for credit risk assessment in digital finance applications.

Feature subset selection is a crucial pre-processing step in building machine learning models, particularly for credit risk assessment. This work addresses the feature selection problem in credit risk modeling using a multi-objective optimization approach. We employ the Multi-Objective Feature Selection - Binary Differential Evolution (MOFS-BDE) algorithm, which simultaneously minimizes classification error and feature subset cardinality. We analyze the existing mutation strategy of MOFS-BDE and identify limitations in its exploration-exploitation balance. To address this, we propose a modified weighted mutation strategy that dynamically adjusts between global exploration and local exploitation across generations. Experimental results on four credit risk datasets (German Credit, Australian Credit, Taiwan Credit, and Credit Card Fraud Detection) demonstrate that the proposed mutation strategy outperforms the original MOFS-BDE regarding minimum error, feature reduction, and hypervolume metrics. Our contributions include: (1) adapting MOFS-BDE for credit risk modeling, (2) analyzing and improving its mutation strategy, and (3) validating the enhanced algorithm on real-world financial datasets.

Credit risk assessment is an essential task for financial institutions, as it evaluates the likelihood of loan defaults. Misclassifying loan applications can result in substantial economic losses. However, accurately predicting customers' creditworthiness presents challenges for machine learning models due to high dimensionality and imbalanced class distributions, which can severely impact overall model performance. This implements three feature selection (FS) techniques, such as Random Forest with Highly Correlated features (RFHC), Information Gain, and Chi-Square to address the high dimensional problem. Additionally, SMOTE is implemented to address the imbalanced dataset. The model performance is analyzed using three different approaches. First, employing only feature selection techniques, in the second approach, we employed feature selection followed by oversampling, and in the final approach, oversampling prior to feature selection is employed. The experimental results indicate that implementing feature selection followed by oversampling gives better performance compared to the other two approaches. Among the feature selection techniques, RFHC outperforms the others, achieving a highest F1-score of $\text{95.12 \%}$ and an AUC value of $\text{85.76 \%}$.

From the early start of the economy, credit risk has been study case for both researchers and financial experts. Financial credit scoring has been the most crucial process in the finance industry as it helps banks and financial institutions to make better decisions. The challenge arises in selecting the features that will make my decision better. This study proposes a feature selection approach using a genetic algorithm based on the information gain to improve classification performance. The genetic algorithm chooses feature subsets trough Logistic Regression. The validation is applied to the Home Credit Default Risk dataset. The experimental results demonstrates that the approach effectively selects features that improves the classification accuracy. The work suggests that evolutionary algorithms combined with mathematical measures can enhance credit risk prediction models and leads to better decision making in financial institutions.

Feature selection is a critical step in credit risk assessment, as redundant and noisy features can significantly degrade model performance. The Dung Beetle Optimizer (DBO), as a novel metaheuristic algorithm, has shown promise in feature selection but still suffers from issues such as susceptibility to local optima and low search efficiency. This paper proposes an Elite-Driven Dung Beetle Optimizer (EDDBO), which enhances the algorithm’s global search and convergence performance through three strategies: elite individuals guidance, elite centroid steering, and elite individuals preservation. By integrating EDDBO with Support Vector Machine (SVM), a feature selection model for credit risk identification is established. Experimental results on nine UCI datasets and four credit risk datasets demonstrate that EDDBO outperforms comparative algorithms in terms of classification accuracy, fitness value, and the number of selected features, thereby validating its effectiveness and superiority in credit risk feature selection.

In finance, predictive models must balance accuracy and interpretability, particularly in credit risk assessment, where model decisions carry material consequences. We present a quantum neural network (QNN) based on a single qudit, in which both data features and trainable parameters are co-encoded within a unified unitary evolution generated by the full Lie algebra. This design explores the entire Hilbert space while enabling interpretability through the magnitudes of the learned coefficients. We benchmark our model on a real-world, imbalanced credit-risk dataset from Taiwan. The proposed QNN consistently outperforms LR and reaches the results of random forest models in macro-F1 score while preserving a transparent correspondence between learned parameters and input feature importance. To quantify the interpretability of the proposed model, we introduce two complementary metrics: (i) the edit distance between the model's feature ranking and that of LR, and (ii) a feature-poisoning test where selected features are replaced with noise. Results indicate that the proposed quantum model achieves competitive performance while offering a tractable path toward interpretable quantum learning.

No abstract available

Large Language Models (LLMs) are increasingly explored as flexible alternatives to classical machine learning models for classification tasks through zero-shot prompting. However, their suitability for structured tabular data remains underexplored, especially in high-stakes financial applications such as financial risk assessment. This study conducts a systematic comparison between zero-shot LLM-based classifiers and LightGBM, a state-of-the-art gradient-boosting model, on a real-world loan default prediction task. We evaluate their predictive performance, analyze feature attributions using SHAP, and assess the reliability of LLM-generated self-explanations. While LLMs are able to identify key financial risk indicators, their feature importance rankings diverge notably from LightGBM, and their self-explanations often fail to align with empirical SHAP attributions. These findings highlight the limitations of LLMs as standalone models for structured financial risk prediction and raise concerns about the trustworthiness of their self-generated explanations. Our results underscore the need for explainability audits, baseline comparisons with interpretable models, and human-in-the-loop oversight when deploying LLMs in risk-sensitive financial environments.

No abstract available

No abstract available

The accuracy of credit risk evaluation is crucial for the profitability of any financial institution. The factorization machine is a widely available model that can effectively be utilized for classification or regression through appropriate feature transformation. In this article, we apply the factorization machine model to the field of credit risk assessment. Since some features of the credit risk assessment data are not numerical, one-hot encoding is used, resulting in sparse training data. However, the computational complexity of the factorization machine is polynomial. To illustrate the effectiveness of the factorization machine credit risk assessment model and compare its performance with other classification approaches such as logical regression, support vector machine, k -nearest neighbors, and artificial neural network, we conduct numerical experiments on four real-world credit risk evaluation datasets. The experimental results demonstrate that the proposed factorization machine credit risk assessment model achieves higher accuracy compared to other machine-learning models on real-world datasets and is computationally more efficient. Therefore, the factorization machine model can be considered as a suitable candidate for credit risk assessment.

Machine learning methods have gained widespread utilization in small and micro enterprise credit risk assessment. However, the practical application of these methods encounters a conundrum involving accuracy and interpretability. In this study, a multi-stage ensemble model is proposed to enhance the model’s interpretability. To strengthen predictive portraits, a multi-feature enhancement method is proposed to integrate non-financial behavioral information and soft information on credit rating into the annual loan ledger data, thereby bolstering the explanatory capacity of the features. To rectify the issue of data imbalance and avoid information loss, a new bagging-based oversampling method is proposed to oversample the minority class samples in multiple parallelized subsets divided by the bagging strategy. To unleash the performance potential of base classifiers, a new voting-weight optimization method is proposed to optimize the soft voting weights of the candidate base classifiers. The experiment results of an annual loan ledger dataset of a commercial bank in China (with an accuracy of 97.9%, an area under the curve of 0.97, a logistic loss of 0.07, a Brier score of 0.01, and a Kolmogorov-Smirnov statistic of 0.38) and the other five public datasets indicating excellent model fit. By focusing on the widespread soft information and data structures characteristic of SME loan risk assessment data, an additional SHAP model explanation method enhances interpretability. This method reveals that the enhanced ‘debt-to-income ratio,’ along with non-financial behavioral information and features derived from soft information, are essential for predicting loan defaults. Such enhancements help to alleviate the issue of information asymmetry in SME loan risk assessment.

Federated Learning (FL) uses local data to perform distributed training on clients and combines resulting models on a public server to mitigate privacy exposure by avoiding data sharing. However, further research indicates that communication overheads continue to be the primary limitation for FL relative to alternative considerations. This is especially true when training models on non-independent and identically distributed data, such as financial default risk data, where FL’s computational costs increase significantly. This study aims to address financial credit risk data by establishing a dynamic receptive field adjustment mechanism for feature extraction, efficiently extracting default features with varying distributions and attributes. Additionally, by constructing a distributed feature fusion architecture, characteristics from both local and overarching models are aggregated to attain higher accuracy with lower transmission costs. Experimental results demonstrate that the proposed FL framework can utilize dynamic receptive fields to allocate convolutional kernel weights, thereby improving feature extraction. In the feature fusion stage, the proposed Multi-Fusion strategy efficiently customizes the aggregation of features from local and global models. The final solution reduces the communication rounds in federated learning by approximately 80%.

No abstract available

No abstract available

Early warnings of enterprise credit risk based on supply chain scenarios are helpful for preventing enterprise credit deterioration and resolving systemic risk. Enterprise credit risk data in the supply chain are characterized by higher‐dimension information and class imbalance. The class imbalance influences the feature selection effect, and the feature subset is closely related to the predictive performance of subsequent learning algorithms. Therefore, ensuring the adaptivity of feature selection and the subsequent class imbalance–oriented classification model is a key issue. We propose an AdaBoost ensemble model with fast nondominated feature selection (AdaFNDFS). AdaFNDFS uses the FNDFS method in the AdaBoost algorithm to iteratively select features and uses the classifier to evaluate the performance of feature subsets to train the class imbalance–oriented classifier and the best‐matched feature subset, ensuring the adaptivity of feature selection and subsequent classifiers. The further use of the differential sampling rate (DSR) method enables AdaFNDFS to integrate more training models with different knowledge and to obtain higher accuracy and better generalization ability for prediction tasks facing high‐dimensional information and class imbalance. A test using credit risk data from Chinese listed enterprises containing supply chain information demonstrates that the prediction scoring indicators, such as AUC, KS, AP, and accuracy, of the AdaFNDFS are better than those of basic models such as LR, LDA, DT, and SVM and multiple hybrid models that use SMOTE, feature selection, and ensemble methods. AdaFNDFS outperforms the basic models by at least 0.0073 (0.0344, 0.0349, and 0.0071) in terms of the AUC (KS, AP, and accuracy). AdaFNDFS has outstanding advantages in predicting enterprise credit risk in the supply chain and can support interested decision‐makers.

Establishing a reasonable and effective feature system is the basis of credit risk early warning. Whether the system design is appropriate directly determines the accuracy of the credit risk evaluation results. In this paper, we proposed a feature system through a validity index with maximum discrimination and commercial banks' loan profit maximization. First, the first objective function is the minimum validity index constructed by the intra-class, between-class, and partition coefficients. The maximum difference between the right income and wrong cost is taken as the second objective function to obtain the optimal feature combination. Second, the feature weights are obtained by calculating the change in profit after deleting each feature with replacement to the sum of all change values. An empirical analysis of 3, 425 listed companies from t-1 to t-5 time windows reveals that five groups of feature systems selected from 614 features can distinguish between defaults and non-defaults. Compared with 14 other models, it is found that the feature systems can provide at least five years' prediction and enable financial institutions to obtain the maximum profit.

With the growth of people’s demand for loans, banks and other financial institutions put forward higher requirements for customer credit risk level classification, the purpose is to make better loan decisions and loan amount allocation and reduce the pre-loan risk. This article proposes a Multi-Level Classification based Ensemble and Feature Extractor (MLCEFE) that incorporates the strengths of sampling, feature extraction, and ensemble classification. MLCEFE uses SMOTE + Tomek links to solve the problem of data imbalance and then uses a deep neural network (DNN), auto-encoder (AE), and principal component analysis (PCA) to transform the original variables into higher-level abstract features for feature extraction. Finally, it combined multiple ensemble learners to improve the effect of personal credit risk multi-classification. During performance evaluation, MLCEFE has shown remarkable results in the multi-classification of personal credit risk compared with other classification methods.

To address the high‐dimensional issues in credit risk assessment, an improved multilayer restricted Boltzmann machine (RBM) based feature extraction method is proposed. In the improved multilayer RBM methodology, the reconstruction error method is first applied to ensure the number of RBM layers to construct an optimal model and then the weighted pruning approach is used to remove redundant and irrelevant traits. For verification purposes, two real‐world credit datasets are employed to demonstrate the effectiveness of the proposed multilayer RBM methodology. The experimental results reveal that a significant improvement in credit classification performance can be obtained by the improved multilayer RBM methodology. This indicates the improved multilayer RBM model proposed in this paper can be used as a promising tool to solve the high‐dimensionality issues in credit risk evaluation.

For the financial health of lenders and institutions, one important risk assessment called credit risk is about correctly deciding whether or not a borrower will fail to repay a loan. It not only helps in the approval or denial of loan applications but also aids in managing the non-performing loan (NPL) trend. In this study, a dataset provided by the LendingClub company based in San Francisco, CA, USA, from 2007 to 2020 consisting of 2,925,492 records and 141 attributes was experimented with. The loan status was categorized as “Good” or “Risk”. To yield highly effective results of credit risk prediction, experiments on credit risk prediction were performed using three widely adopted supervised machine learning techniques: logistic regression, random forest, and gradient boosting. In addition, to solve the imbalanced data problem, three sampling algorithms, including under-sampling, over-sampling, and combined sampling, were employed. The results show that the gradient boosting technique achieves nearly perfect Accuracy, Precision, Recall, and F1score values, which are better than 99.92%, but its MCC values are greater than 99.77%. Three imbalanced data handling approaches can enhance the model performance of models trained by three algorithms. Moreover, the experiment of reducing the number of features based on mutual information calculation revealed slightly decreasing performance for 50 data features with Accuracy values greater than 99.86%. For 25 data features, which is the smallest size, the random forest supervised model yielded 99.15% Accuracy. Both sampling strategies and feature selection help to improve the supervised model for accurately predicting credit risk, which may be beneficial in the lending business.

Aiming at the problems of insufficient feature discrimination, difficulty in parameter adjustment, and lack of time dependence capture ability in the existing credit risk prediction models of listed companies, this paper proposes a PSOCNN-BiLSTM-MQA credit risk prediction method PCBM based on hierarchical feature modeling. Based on the financial data of listed companies in Guotai’an database, the method divides the data into four types of first-level characteristics: debt paying ability, operating ability, profitability and development ability. Each class of first-level features is further subdivided into several second-level features. In the feature extraction stage, a twochannel Convolutional Neural Network (CNN) was used to extract the local features of each class of level-1 features. One channel processed the original data, and the other channel processed the features converted to grayscale images. After being processed by different CNNS, the multi-query attention mechanism (MQA) was used to fuse the features from different channels to enhance the attention to features at different levels. At the same time, the bidirectional long short-term memory network is used to capture the long-term dependencies in the time series data, and the fused features are integrated through the MQA mechanism. In order to further improve the performance of the model, the particle swarm optimization algorithm is used to optimize the model parameters. Experimental results show that compared with the existing prediction models, the PCBM model shows significant improvement in accuracy, precision, recall rate and F1-score, which proves the effectiveness of this method in the credit risk prediction of listed companies.

In the digital financial services era, Peer-to-Peer (P2P) lending has emerged as a significant innovation in fintech. However, credit risk remains a major concern due to the potential for payment failures, which can cause losses for platforms and investors. This research explores the impact of Deep Feature Synthesis (DFS) on credit risk classification and evaluates the performance of the Light Gradient Boosting Machine (LightGBM) algorithm with and without DFS. The data used in this study was sourced from Kaggle, a peer-to-peer lending company based in San Francisco, California, United States. The dataset contains 74 attributes, with a total of 887,379 rows. DFS automatically generates new attributes, while LightGBM is used for selecting the most important features, aiming to optimize credit risk predictions and simplify the model's complexity. The results of credit risk classification models using DFS and without it. Findings reveal that DFS enhances the accuracy of the credit risk classification, achieving a 0.99 accuracy rate compared to 0.97 without DFS, achieving a recall and F1-score of 0.94 and 0.96 with DFS and 0.68 and 0.81 without DFS. These results suggest that DFS is an effective feature engineering technique for boosting credit risk model performance. This research contributes significantly to the P2P lending industry by demonstrating that combining DFS with LightGBM can improve credit risk management, making it a valuable approach for financial platforms.

No abstract available

ABSTRACT Lending platforms operating on a peer-to-peer (P2P) basis encounter the intricate challenge of assessing borrower creditworthiness to minimize the risk of defaults. This study addresses this challenge by proposing an advanced approach to feature selection that leverages the Grey Wolf Optimizer (GWO) in conjunction with a finely tuned Decision Tree (DT) model. The main objective is to enhance the precision and efficiency of feature selection processes within P2P lending datasets. The study begins by fine-tuning DT hyperparameters using Genetic Algorithms (GA), yielding an optimal configuration: ‘max_depth’ = 40, ‘min_samples_leaf’ = 20, and ‘criterion’ = ‘entropy’. Subsequent phases involve the application of GWO and modified GWO (nGWO, cGWO, and lGWO) to conduct feature selection under distinct Search Agent (SA) setups (SA = 5, SA = 20, SA = 50). Particularly impressive is the performance of the lGWO model with the SA = 50 setup, achieving a remarkable 91% accuracy while selecting 80.55% of the total 36 features. This study significantly improves how lenders manage risks in P2P lending by identifying high-risk borrowers more effectively, helping lenders reduce financial risks and benefiting all parties involved.

Diabetes is a chronic disease that affects a significant portion of the global population. It occurs when the body cannot produce enough insulin or effectively use the insulin it produces, leading to elevated blood glucose levels. Diabetes is a major contributor to various severe health conditions, including heart disease, stroke, kidney failure, and nerve damage. Early detection of diabetes is crucial in mitigating these associated health risks and improving patient outcomes. In response to the increasing prevalence of diabetes, we have developed an automated system for Early-Stage Diabetes Risk Prediction (ESDRP). This study utilizes a dataset consisting of 16 features from 520 instances. We applied multiple Machine Learning (ML) models, including XGBoost (XGB), Bootstrap Aggregating (BAG), Adaptive Boosting (AdaBoost), Light Gradient Boosted Machine (LGBM), and Gradient Boosting Decision Trees (GBDT), both with and without feature generation techniques. Specifically, we explored polynomial and binning feature generation methods. Our findings indicate that the polynomial feature generation technique combined with XGB yielded the highest performance, achieving an accuracy of 99.22%, precision of 100%, recall of 98.15%, specificity of 99.06%, and F1-score of 100%. Additionally, all the ML models were evaluated using confusion matrices (CM) and ROC curves, with the average performance across 10-fold cross-validation demonstrating robust predictive capabilities. Furthermore, to establish trust in our model's predictions, we incorporated two explainable AI (XAI) methods: LIME and SHAP. These techniques helped us understand feature importance, the decision-making process of the models, and enhanced the reliability of our results. Our automated system aims to assist individuals of all ages and healthcare systems in identifying ESDRP, thereby supporting informed decision-making and improving global health outcomes.

No abstract available

In classification, feature selection engineering helps in choosing the most relevant data attributes to learn from. It determines the set of features to be rejected, supposing their low contribution in discriminating the labels. The effectiveness of a classifier passes mainly through the set of selected features. In this paper, we identify the best features to learn from in the context of credit risk assessment in the financial industry. Financial institutions concur with the risk of approving the loan request of a customer who may default later, or rejecting the request of a customer who can abide by their debt without default. We propose a feature selection engineering approach to identify the main features to refer to in assessing the risk of a loan request. We use different feature selection methods including univariate feature selection (UFS), recursive feature elimination (RFE), feature importance using decision trees (FIDT), and the information value (IV). We implement two variants of the XGBoost classifier on the open data set provided by the Lending Club platform to evaluate and compare the performance of different feature selection methods. The research shows that the most relevant features are found by the four feature selection techniques.

The assessment of credit risk for P2P lending platform applicants is critical to investors. Feature engineering is an essential technique in distilling classification knowledge during the credit risk prediction data preprocessing stage. Although previous literature used feature selection methods to identify key features, feature transformation is more useful in discovering intrinsic nonlinear characteristics in credit data. In this study, we propose a synthetic multiple tree‐based feature transformation method to generate features. Multiple tree‐based feature transformation methods are employed and fused to acquire a new feature set. The bagging‐based tree ensemble feature transformation method (Bagging‐TreeEnsembleFT) and boosting‐based tree ensemble feature transformation method (Boosting‐TreeEnsembleFT) are two types of feature transformation methods that we specifically propose to validate their effect. We verify the credit risk prediction performance using the proposed synthetic feature transformation methods on real P2P Lending credit datasets. Empirical analysis demonstrates that tree‐based ensemble feature transformation methods with boosting ensemble strategy achieve better prediction performance on various datasets corresponding to different partitions and class distributions compared to tree‐based ensemble feature transformation methods with bagging ensemble strategy and individuals. Moreover, the proposed synthetic feature transformation method improves the credit risk prediction performance in terms of accuracy, AUC, and F1‐score.

Outlier detection is currently applied in many fields, where existing research focuses on improving imbalanced data or enhancing classification accuracy. In the financial area, financial fraud detection puts higher demands on real-time and interpretability. This paper attempts to develop a credit risk model for financial fraud detection based on an extreme gradient boosting tree (XGBoost). SMOTE is adopted to deal with imbalanced data. AUC is the assessment indicator, and the running time is taken as the reference to compare with other frequently used classification algorithms. The results indicate that the method proposed by this paper performs better than others. At the same time, XGBoost can obtain a ranking of important features that impact the classification results when performing classification tasks, making the evaluation results of the model interpretable. The above shows that the model proposed in the paper is more practical in solving credit risk assessment problems. It has faster response times, reduced costs, and better interpretability.

This study presents a gradient-informed proxy initialization framework designed to improve training efficiency and predictive performance in deep learning models for time-series forecasting. The method extends the Laor Initialization approach by introducing backward gradient norm clustering as a selection criterion for input-layer weights, evaluated through a lightweight, architecture-agnostic proxy model. Only the numerical input layer adopts the selected initialization, while internal components retain standard schemes such as Xavier, Kaiming, or Orthogonal, maintaining compatibility and reducing overhead. The framework is evaluated on a real-world financial forecasting task: identifying high-risk equities from the Thai Market Surveillance Measure List, a domain characterized by label imbalance, non-stationarity, and limited data volume. Experiments across five architectures, including Transformer, ConvTran, and MMAGRU-FCN, show that the proposed strategy improves convergence speed and classification accuracy, particularly in deeper and hybrid models. Results in recurrent-based models are competitive but less pronounced. These findings support the method’s practical utility and generalizability for forecasting tasks under real-world constraints.

In many practical applications, such as fraud detection, credit risk modeling or medical decision making, classification models for assigning instances to a predefined set of classes are required to be both precise as well as interpretable. Linear modeling methods such as logistic regression are often adopted, since they offer an acceptable balance between precision and interpretability. Linear methods, however, are not well equipped to handle categorical predictors with high-cardinality or to exploit non-linear relations in the data. As a solution, data preprocessing methods such as weight-of-evidence are typically used for transforming the predictors. The binning procedure that underlies the weight-of-evidence approach, however, has been little researched and typically relies on ad-hoc or expert driven procedures. The objective in this paper, therefore, is to propose a formalized, data-driven and powerful method. To this end, we explore the discretization of continuous variables through the binning of spline functions, which allows for capturing non-linear effects in the predictor variables and yields highly interpretable predictors taking only a small number of discrete values. Moreover, we extend upon the weight-of-evidence approach and propose to estimate the proportions using shrinkage estimators. Together, this offers an improved ability to exploit both non-linear and categorical predictors for achieving increased classification precision, while maintaining interpretability of the resulting model and decreasing the risk of overfitting. We present the results of a series of experiments in a fraud detection setting, which illustrate the effectiveness of the presented approach. We facilitate reproduction of the presented results and adoption of the proposed approaches by providing both the dataset and the code for implementing the experiments and the presented approach.

No abstract available

In order to make up for the shortage of public-private partnership (PPP) model, more and more sewage treatment PPP projects have adopted the asset-backed securitization (ABS) model. To ensure success of sewage treatment PPPABS projects, risk evaluation, which has remained scarcity and unscientific, is becoming an urgent problem to be solved. Firstly, this paper identifies critical risk factors by literature analysis and expert interview. The final risk system is established from the perspectives of macrorisks, basic asset risks, transaction structure risks, operational risks, and other risks, which include 17 second risk factors. Then, the overall risk evaluation method is proposed based on combination weight method and Dempster–Shafer (D-S) evidence theory. Next, Beijing capital Co. Ltd. sewage treatment PPPABS project as a case is employed to verify the feasibility and effectiveness of the proposed method. Finally, awareness of existing risks, suggestions from law risk, quality risk, underwriting and issue risk, and credit enhancement are provided for sewage treatment PPPABS projects. All above studies are expected to provide helpful references for evaluating overall risk of sewage treatment PPPABS projects.

No abstract available

A method based on improved fuzzy theory of evidence was presented to solve the problem that there exist all kinds of uncertainty in the process of information security risk assessment. The hierarchy model for the information systems risk assessment was established firstly, and then fuzzy sets were introduced into theory of evidence. The basic probability assignments were constructed using the membership function of fuzzy sets, and the basic probability assignments were determined. Moreover, weight coefficients were calculated using entropy weight and empirical factor, which combined the objective weights with the subjective ones, and improved the validity and reliability. An illustration example indicates that the method is feasible and effective, and provides reasonable data for constituting the risk control strategy of the information systems security.

No abstract available

This workshop aimed to elucidate the contribution of computational and emerging in vitro methods to the weight of evidence used by risk assessors in food safety assessments. The following issues were discussed: using in silico and high-throughput screening (HTS) data to confirm the safety of approved food ingredients, applying in silico and HTS data in the process of assessing the safety of a new food ingredient, and utilizing in silico and HTS data in communicating the safety of food ingredients while enhancing the public’s trust in the food supply. Perspectives on integrating computational modeling and HTS assays as well as recommendations for optimizing predictive methods for risk assessment were also provided. Given the need to act quickly or proceed cautiously as new data emerge, this workshop also focused on effectively identifying a path forward in communicating in silico and in vitro data.

No abstract available

No abstract available

No abstract available

Investment location decisions in Southeast Asia face complex challenges involving multidimensional economic indicators, geopolitical risks, and multi-level spatial choices. This study proposes the Southeast Asia Risk Perception Meta-Strategy Optimization Algorithm (SEAR-MPO), constructing a reinforcement learning decision-making framework that integrates multimodal state encoding and hierarchical action space decomposition. The algorithm integrates heterogeneous features such as overseas Chinese population density, geographical distance, macroeconomic factors, and political risks through a spatial-economic joint encoder, and designs a risk-triggered hybrid update mechanism to dynamically adjust the weight configuration between Q-learning and SARSA. Empirical analysis based on the 2015-2024 dataset of six Southeast Asian countries shows that the SEAR-MPO model significantly outperforms traditional benchmark models in terms of return rate, Sharpe ratio, and volatility. Location selection identified the best investment destinations, with VSIP and Ho Chi Minh leading with annualized returns of 13% and 12%, respectively, reflecting the strong growth potential of the Vietnamese market. It also indicates that cultural synergy drives capital to cluster in areas with high concentrations of overseas Chinese, particularly Singapore, which highlights its position of cultural convergence. This study provides an interpretable, risk-adaptive intelligent decision-making paradigm for dynamic investment decisions in emerging markets.

In this work, we study the weighted empirical risk minimization (weighted ERM) schema, in which an additional data-dependent weight function is incorporated when the empirical risk function is being minimized. We show that under a general ``balanceable"Bernstein condition, one can design a weighted ERM estimator to achieve superior performance in certain sub-regions over the one obtained from standard ERM, and the superiority manifests itself through a data-dependent constant term in the error bound. These sub-regions correspond to large-margin ones in classification settings and low-variance ones in heteroscedastic regression settings, respectively. Our findings are supported by evidence from synthetic data experiments.

Recent increases in overweight and obesity have established Metabolic–Bariatric Surgery (MBS) as a principal intervention for durable weight reduction and metabolic improvement. Given the elevated perioperative complication risk among patients with obesity, there is a growing imperative to enhance the precision of preoperative management. This review synthesizes evidence from 2020–2025 on the application of artificial intelligence (AI) to bariatric surgery preoperative assessment, focusing on machine learning (ML), deep learning (DL), and large language models (LLMs). We summarize AI applications across three domains: preoperative risk prediction and individual assessment, patient stratification and procedure selection, and preoperative education and surgical training. The evidence suggests that integrating AI into routine preoperative workflows enables individualized, quantitative estimation of high-risk complications and weight-loss prognosis, thereby optimizing risk management; it can also support patient stratification and procedure matching to facilitate patient-centered shared decision-making; and NLP- and vision-based tools promote standardization and visualization of knowledge for patients and surgeons. Overall, AI is driving preoperative assessment toward greater individualization, interpretability, and multidisciplinary coordination, but its clinical generalizability requires validation in multicenter, prospective studies.

The interdependence inherent in interbank networks amplifies vulnerability to systemic risk, particularly through correlated asset exposures during exogenous negative shocks. This study employs exponential random graph models (ERGMs) to reconstruct a bipartite network of asset-holding correlations based on the balance sheets of Chinese commercial banks from 2016 to 2022. The reconstructed network closely approximates the topological features of the actual banking system. We then introduce a novel framework for measuring aggregate network vulnerability, which incorporates bank size, initial shocks, interconnectedness, leverage, and asset fire sales to capture key channels of financial contagion. Our results indicate that the reconstructed network aligns closely with empirical data in both link structure and weight distribution. Furthermore, cumulative systemic vulnerability increases non-linearly with the severity of the initial shock and the discount depth of fire sales. For individual banks, indirect vulnerability driven by contagion via deleveraging and fire sales significantly exceeds direct losses from initial shocks. Systemic risk contributions are concentrated in large state-owned banks and nationwide joint-stock commercial banks, whereas the institutions most susceptible to risk shocks are predominantly small and medium-sized rural and urban commercial banks.

Plant diseases can severely impact crop yields, posing a major risk to worldwide food stability. Prompt and precise identification of these diseases is crucial for early intervention and efficient crop administration. This paper introduces an innovative method for detecting plant leaf diseases using residual networks (ResNets) and the PlantVillage dataset. To develop light weight residual (LWR) architecture, five convolutional layers are interleaved with five max-pooling layers, making up the architecture of ten layers. The number of filters in the convolutional layers is gradually increased from 32 to 64 and up to 512 with a 3×3 kernel. A fully connected layer is the last layer of the network which provides the classification of leaf diseases The LWR architecture is trained and evaluated using the PlantVillage dataset, a broad collection of annotated images. This dataset serves as the basis for the system. The findings of the experiments provide evidence that the suggested system has higher accuracy, sensitivity, and specificity measures. The use of residual networks in LWR architecture improves the capability of the model to acquire complicated representations, which in turn enables a more precise differentiation between healthy and unhealthy plant leaves.

This paper presents a detailed analysis of the Holt-Winters-GRU hybrid model for predicting global rice prices, an essential agricultural commodity. The benefits of the traditional statistical approaches are combined with deep learning power, and the results have been found to outperform a standalone GRU. The hybrid model produced a test RMSE of 27.7532 with almost no difference between the training and testing errors, thus showing robust generalization ability. Detailed scrutiny of the weight heat map for the GRU layer reflects the intricacies of the model while depicting both seasonal patterns and intricate nonlinear relationships present in the rice price time series. The findings from the study reveal that the Holt-Winters-GRU hybrid model is usable in forecasting rice price movements for policymakers, traders, and market analysts, considering its ability to handle systematic trends and shocks. Recommendations for model implementation, enhancement, risk management, policy applications, and future research are provided to extend further the utility of this hybrid forecasting approach in agricultural commodity markets. 

Belief divergence is a significant measure to quantify the discrepancy between evidence, which is beneficial for conflict information management in Dempster–Shafer evidence theory. In this article, three new concepts are given, namely, the belief Bhattacharyya coefficient, adjustment function, and enhancement factor. And based on them, a novel enhanced belief divergence, called EBD, is proposed, which can assess the correlation of subsets and fully reflect the uncertainty of multielement sets. The important properties of the EBD have been studied. In particular, a new EBD-based multisource information fusion method is designed to handle evidence conflict, where the weight of evidence is decided by the EBD between evidence and the information volume of each evidence. Compared with other methods, the proposed method in the applications of target recognition and iris classification can produce more rational and telling outcomes when dealing with conflict information. Finally, an application in risk priority evaluation of the failure modes of rotor blades of an aircraft turbine is provided to validate that the proposed method has the extensive applicability.

Despite being the most prevalent complication, cardiovascular risk factors such as blood pressure, weight, and lipid profile have been less considered in digital health studies. The aim of this systematic review and meta-analysis was to gather evidence regarding the impact of digital health applications on cardiovascular risk factors in patients with diabetes. Literature search was conducted following the PRISMA guideline on September 4, 2023, using databases including PubMed, Scilit, Scopus, Embase, and Web of Science, with a pre-planned combination of keywords. Selected studies were original research reporting the influence of smartphone applications on cardiovascular risk factors in diabetic patients. Standardized mean differences (SMD) between the intervention and control groups were analyzed using fixed or random-effects models. Eighteen studies met the criteria, consisting of 1152 patients in the intervention group and 1072 patients in the control group. The results of the meta-analysis showed that the smartphone applications significantly controlled systolic blood pressure (SMD: -5.03 mmHg; 95%CI: -7.018, -3.041, p<0.001). There was no significance effect on weight, body mass index, total cholesterol, low-density lipoprotein cholesterol (LDL-c), high-density lipoprotein cholesterol (HDL-c) and diastolic blood pressure. In the subgroup analysis, triglycerides were lower in the intervention group compared to the control group (SMD: -0.459%; 95%CI: -0.787, -0.132, p=0.006). Publication bias and the limited number of studies suggest that the evidence from this study is in moderate level. In conclusion, smartphone apps are not only effective in aiding blood sugar control but also in preventing cardiovascular issues in diabetic patients. Further research is still needed to confirm these findings.

Jiangxi Province, characterized by abundant forest resources and complex topography, is highly susceptible to forest fires. This study integrated multiple factors, including topography, climate, vegetation, and human activities, and employed machine learning models, specifically random forest (RF), support vector machine (SVM), and back-propagation neural network (BPNN), to predict forest fire occurrence in Jiangxi. Using Moderate Resolution Imaging Spectroradiometer L3 fire-point data from 2001–2020, we analyzed the spatiotemporal distribution of forest fires and applied the weight of evidence (WoE) method to evaluate the correlation between forest fires and environmental factors. WoE was employed to select negative samples, which were compared with those obtained using traditional random sampling methods. The optimal model was then utilized to generate seasonal spatial distribution maps of forest fire risk throughout Jiangxi Province. The results showed that over the past two decades, the frequency of forest fires generally decreased. RF demonstrated a significant advantage over SVM and BPNN in predicting forest fires. Vegetation coverage was the most influential factor. In addition, the models trained with WoE-selected negative samples exhibited enhanced accuracy, with area under the curve values increasing from 0.946 to 0.995 for RF, 0.8344 to 0.925 for SVM, and 0.832 to 0.850 for BPNN, compared to those trained with randomly sampled negative data. Finally, forest fires were most frequent during winter, particularly in Ganzhou, Fuzhou, and Ji'an. High-risk fire zones were more dispersed in spring, whereas autumn fires were primarily concentrated in Ganzhou, and fire activity was relatively low during summer. The seasonal forest fire risk maps generated in this study offer valuable insights for guiding forest fire management in the Jiangxi Province and similar regions, providing critical practical significance for informed decision-making.

Abstract. Effective risk management is vital for ensuring the stability and profitability of financial institutions. This study focuses on enhancing credit risk assessment by developing a scoring model that quantifies customer risk factors using logistic regression. Key features such as Weight of Evidence (WOE) and Information Value (IV) were employed to transform and select variables. The datasets used in this research, cs-training.csv and cs-test.csv, were preprocessed, including handling missing data and binning continuous variables to improve model interpretability and performance. The logistic regression model yielded an AUC value of 0.78, demonstrating strong predictive capabilities, though optimization is necessary to improve the F1 score. The final scorecard generated from the model provides actionable insights for financial institutions, enabling more accurate risk predictions and decision-making. This tool is particularly useful for loan approvals and credit issuance, offering a data-driven approach to managing credit risk in modern financial environments.

No abstract available

No abstract available

Security problem in WLAN has always been a hot topic, but traditional assessment methods are often not suitable for WLAN. By deeply analyzing the typical characteristics and security attributes of WLAN, a hierarchical security risk assessment model is established. First analytic hierarchy process is used to determine weight of risk assessment index in the model, and then improved D-S combination rules of evidence theory is used to fuse risk assessment indexes from bottom to top, and eliminate the conflict between different evaluation values, next final risk value of WLAN is obtained with a comprehensive consideration of various factors including asset value, vulnerability severity, and threat frequency. Finally simulation results verify feasibility and validity of the proposed method, which can provide objective basis of decision and pointed guidance for security management of WLAN.

Systems that answer questions by reviewing the scientific literature are becoming increasingly feasible. To draw reliable conclusions, these systems should take into account the quality of available evidence from different studies, placing more weight on studies that use a valid methodology. We present a benchmark for measuring the methodological strength of biomedical papers, drawing on the risk-of-bias framework used for systematic reviews. Derived from over 500 biomedical studies, the three benchmark tasks encompass expert reviewers'judgments of studies'research methodologies, including the assessments of risk of bias within these studies. The benchmark contains a human-validated annotation pipeline for fine-grained alignment of reviewers'judgments with research paper sentences. Our analyses show that large language models'reasoning and retrieval capabilities impact their effectiveness with risk-of-bias assessment. The dataset is available at https://github.com/RoBBR-Benchmark/RoBBR.

ABSTRACT Increasing global interest in diet and fitness mobile applications (apps) has prompted the question: What are the factors affecting users’ adoption and usage behaviors on a specific fitness app? By combining the unified theory of acceptance and use of technology (UTAUT) with the health belief model (HBM), and including risk perception of information technology with the farsighted planner and myopic doer from the theory of self-control, we explore the understanding of this academic question. We analyzed data from 8,840 users of Boohee, a diet and fitness app (ranked first in the weight-loss category on the App Store in China). Structural equation modeling revealed that self-efficacy as well as the perceived benefits, barriers, and threats of weight loss significantly influence a fitness app’s performance expectancy, which, in turn, predicts users’ intention to adopt it. Furthermore, actual usage behavior (i.e., diet, exercise, weight, and login records within 30 days after respondents completed the questionnaire) is positively affected by weight-loss intention and behavioral intention to use the app and negatively affected by users’ risk perception. The main findings of this research could help healthcare practitioners and app developers find better ways to encourage people to adopt health apps for various reasons. App developers should attach more importance to users’ actual continuous use behavior than to their intention to use an app. They should provide sufficient introductory information about their apps, thereby reducing users’ risk perception and generating reasonable performance expectancy of the app, so as to improve users’ actual continuous use behavior.

Failure mode and effects analysis (FMEA) has been widely used for potential risk modeling and management. Expert evaluation is used to model the risk priority number to determine the risk level of different failure modes. Dempster–Shafer (D–S) evidence theory is an effective method for uncertain information modeling and has been adopted to address the uncertainty in FMEA. How to deal with conflicting evidence from different experts is an open issue. At the same time, different professional backgrounds of experts may lead to different weights in modeling the evaluation. How to model the relative weight of an expert is an important problem. We propose an improved risk analysis method based on triangular fuzzy numbers, the negation of basic probability assignment (BPA) and the evidence distance in the frame of D–S evidence theory. First, we summarize and organize the expert’s risk analysis results. Then, we model the expert’s assessments based on the triangular fuzzy numbers as BPAs and calculate the negation of BPAs. Third, we model the weight of expert based on the evidence distance in the evidence theory. Finally, the Murphy’s combination rule is used to fuse the risk assessment results of different experts and calculate the new risk priority number (RPN). At the end of this paper, we apply the proposed method to analyze seventeen failure modes of aircraft turbine blades. The experimental results verify the rationality and effectiveness of this method.

Abstract Multiple attribute decision making (MADM) problems often consists of quantitative and qualitative attributes which can be assessed by numerical values and subjective judgments. Subjective judgments can be evaluated by linguistic variables, and both numerical values and subjective judgments can be accurate or uncertain. The evidential reasoning (ER) approach provides a process for dealing with MADM problems of both a quantitative and qualitative nature under uncertainty. The existing ER approach considers both benefit and cost attributes in the evidence combination process. In this paper, deviated interval and fixed interval attributes are introduced into ER based MADM approach and the frames of discernment for representing these two kinds of attributes are given. The transformation rules from the assessment values of deviated interval attributes to belief degrees in the ER structure are then studied. An ave-entropy based weight assignment method considering the risk preference of decision maker is also shown to deal with uncertain assessment situation, such as belief distribution with qualitative attribute and uncertain utility function. Some programming models to generate interval weights and utilities are constructed. The rationality and efficiency of the methods in supporting MADM problems are discussed. Two case studies are provided to demonstrate the applicability and validity of the proposed approaches and the potential in supporting MADM under uncertainty.

Credit risk prediction, reliability, monitoring and effective loan processing are the keys to proper bank decision-making. So, understanding the credit customer during the initial loan processing phase would help the bank prevent future losses. In this regard, this study aims to develop a credit risk evaluation model using deep learning algorithms. The model utilizes a credit risk analysis dataset published in Kaggle. The objective is to build deep learning models for predicting credit risk using real banking datasets published on Kaggle. Firstly, data preprocessing and feature engineering are done. Suitable features such as irrelevant and null valued features are identified and removed with techniques like the Karl Pearson correlation, information values, and weight of evidence. Next, data normalization is performed and target features are separated into three classes: high risk, medium risk and low risk. SMOTE-ENN (Synthetic Minority Oversampling Technique with Edited Nearest Neighbor) was applied to balance the dataset. State-of-the-art deep learning algorithms such as GRU (Gated Recurrent Units) Model and Bidirectional Long Short-Term Memory (Bi-LSTM) are implemented to train and learn from the pre-processed data. GRU and Bi-LSTM models performed well, with F1 scores of 0.92 and 0.93, respectively. The result of this investigation illustrates that deep learning models seem promising for evaluating and predicting multi-class problems.

This survey by the European Centre for Ecotoxicology and Toxicology of Chemicals (ECETOC) highlights that 'omics technologies are generally not yet applied to meet standard information requirements during regulatory hazard assessment. While they are used within weight-of-evidence approaches to investigate substances' modes-of-action, consistent approaches for the generation, processing and interpretation of 'omics data are not applied. To date, no 'omics technology has been standardised or validated. Best practices for performing 'omics studies for regulatory purposes (e.g., microarrays for transcriptome profiling) remain to be established. Therefore, three frameworks for (i) establishing a Good-Laboratory Practice-like context for collecting, storing and curating 'omics data; (ii) 'omics data processing; and (iii) quantitative WoE approaches to interpret 'omics data have been developed, that are presented in this journal supplement. Application of the frameworks will enable between-study comparison of results, which will facilitate the regulatory applicability of 'omics data. The frameworks do not constitute prescriptive protocols precluding any other data analysis method, but provide a baseline for analysis that can be applied to all data allowing ready cross-comparison. Data analysis that does not follow the frameworks can be justified and the resulting data can be compared with the Framework-based common analysis output.

Traditional qualitative risk identification methods, which rely heavily on manual judgment, are increasingly inadequate for modern auditing due to the complexity of business operations and the data-intensive nature of audit scenarios. To address this, this paper proposes a deep learning-based framework for the automated calculation of audit risk profiling indicators. The method first employs semantic parsing to process multi-source unstructured text, extracting entity relationships and constructing feature vectors. Leveraging the reasoning capabilities of deep learning, these semantic segments are then mapped to operational risk dimensions to form a reusable tagging system. Crucially, the framework transforms qualitative descriptions into quantifiable indicators through a multi-dimensional scoring mechanism that evaluates semantic confidence, risk severity, and compliance deviation. Empirical results demonstrate that this approach significantly enhances the coverage, accuracy, and interpretability of risk identification. By achieving automated monitoring and evaluation, this study provides an effective pathway for the digital and intelligent transformation of internal enterprise auditing.

In recent years, automated driving technology is booming and reshaping the pattern of transportation, but its safety has attracted wide attention. In the embedded system, it is necessary to solve the security risks caused by sensor failures, complex road environments and emergencies. To enhance autonomous driving safety, risk detection, a crucial part, must accurately spot potential hazards. This review focuses on the risk detection and safe driving field of autonomous driving, and analyzes related cutting-edge technologies in depth. In terms of risk prediction, it explains the key technology paths based on multi-sensor fusion, deep learning, and data-driven, and realizes risk judgment and insight from comprehensive information capture, deep feature mining, and model optimization, respectively. Safety driving technology section, discusses lane keeping and departure warning, collision warning and automatic braking, blind area monitoring and obstacle avoidance technology. In the future, deep learning, edge calculation, multi-sensor fusion will promote autonomous driving to break through the dilemma, but challenges such as regulations, ethics, and network security will also follow.

This research proposes a state-of-the-art method of automated lung cancer risk prediction using the union of convolutional neural networks (CNNs) and radiomic feature extraction from low dose computed tomography (LDCT) scans. From the publicly accessible LIDC-IDRI dataset containing over 1,000 labeled LDCT scans having rich data about lung nodules, this study isolates quantitative radiomic features such as texture, shape, intensity, and edge sharpness employing software like 3D Slicer and Pyradiomics. Four deep CNN models—VGG16, VGG19, ResNet50, and Inception V3 are trained and validated with TensorFlow and PyTorch platforms. The models are differentiated on accuracy, precision, recall, and F1-score. Inception V3 is the highest in accuracy of 98.67%, then ResNet50 (94.50%), VGG19 (90.23%), and VGG16 (88.56%). The confusion matrices also validate the robustness of each model to differentiate between lung cancer stages. Radiomic feature integration through handcrafted and deep learning have significantly improved classification accuracy, enabling a strong framework for early lung cancer detection and risk stratification. The hybrid method has clinical translation potential through enhanced diagnosis accuracy and patient-specific treatment planning. The study concludes that AI-based methods of this type have the potential to be effective tools in modern medical imaging and cancer diagnosis pipelines.

Organizations implementing machine learning in regulated environments face critical challenges in maintaining transparency, explainability, and compliance as automated decision-making proliferates across financial services, healthcare, and retail sectors. This paper presents a comprehensive framework addressing these challenges through three integrated components: a unified metadata system capturing complete decision context, a scalable feature store architecture supporting dual-mode access patterns, and transparent risk scoring mechanisms generating human-interpretable explanations. The proposed architecture enables intelligent risk scoring systems that balance high performance with regulatory compliance through versioned feature repositories, structured lifecycle management, and continuous learning capabilities. Novel contributions include: (1) unified metadata architecture enabling sub-second lineage queries through graph-based navigation, (2) dual-mode feature store eliminating train-serve skew via synchronized batch and streaming interfaces, and (3) interpretable risk scoring combining SHAP-based attribution with automated explanation generation for regulatory compliance. Implementation across three financial institutions demonstrates measurable improvements in decision traceability, model stability, and operational efficiency while preserving the agility essential for effective machine learning deployments in regulated domains.

Maternal health is a significant global crisis that affects women every day, resulting in severe complications related to pregnancy or childbirth. As a response to this urgent need for effective risk management, we have developed an automated system for maternal health risk (MHR) prediction that utilizes machine learning (ML) and explainable artificial intelligence (XAI). Our research aims to enhance the accuracy and efficiency of risk assessment by employing rigorous preprocessing techniques on a dataset of 1014 samples. To achieve this goal, we employ 10 ensemble ML models and use diverse feature optimization methods such as principal component analysis (PCA), linear discriminant analysis (LDA), and recursive feature elimination (RFE). To clearly understand the decision‐making processes of the selected ensemble model, we employ XAI as Shapley additive explanations (SHAP) plots, local interpretable model‐agnostic explanation (LIME) plots, and individual conditional expectation (ICE) plots. Our evaluation includes various ML performance metrics and a variety of statistical measures, which demonstrate that the Bagging, in conjunction with PCA, emerges as the optimal model, achieving an impressive accuracy of 99.51%. This research uses 10‐fold cross‐validation throughout the whole analysis. By emphasizing proactive risk detection and personalized interventions, we aim to improve understanding of MHR and help women worldwide make informed decisions.

ABSTRACT An arrhythmia is just a disease associated with irregular heart rhythm. Any kind of arrhythmia can be determined with an ECG test. However, this requires long and excellent-quality of ECG samples. This makes an affordable, reliable, and automated system for recognizing and diagnosing arrhythmias. To meet the aforementioned criteria, a new technique with the aid of an attention-aided deep learning approach is proposed to classify arrhythmia in this paper. “Higher-order statistical (HOS)” features are extracted from the original EEG signal and the deep features are obtained from the analyzed QRS complex signals and RR interval by means of a “One-Dimensional Convolution Neural Network (1DCNN)”. The Risk Rate Enhanced Lemurs Optimization Algorithm (RRE-LOA) is utilized to modify the weight, and the extracted features are selected optimally before carrying out the feature selection followed by feature concatenation with weight fusion in order to generate the Weighted Fused Features (WFF). A range of indicators is used to compare the performance of the executed classification model with the traditional models. The results demonstrate that the proposed methodology for classifying arrhythmias offers an enhanced performance while classifying the arrhythmia thus aiding in faster treatment of the arrhythmia disease.

Abstract— This paper proposes a hybrid ML/DL approach for automating compliance risk register generation. A stacked generalization model is built using Random Forest, Logistic Regression, and MLP as base learners, combined via a meta-learner to improve risk classification. The system is trained on synthetic compliance data and utilizes encoding and normalization for preprocessing. Integrated with a Streamlit dashboard, the model provides real-time risk prediction with confidence scores, enabling efficient and accurate compliance risk management. Experimental results demonstrate superior performance compared to individual models, achieving 94.2% accuracy in risk categorization with reduced false positive rates. The system incorporates automated risk scoring mechanisms that adapt to evolving compliance landscapes, while maintaining interpretability through feature importance analysis and uncertainty quantification. Additionally, the framework supports multi-jurisdictional compliance requirements and provides automated reporting capabilities for audit trails and regulatory documentation. The ensemble approach leverages the complementary strengths of traditional machine learning and deep learning techniques to enhance prediction robustness across diverse risk scenarios. Keywords— Risk Register, Stacked Generalization, Stacked Generalization, Compliance Risk Classification, Synthetic Data, Machine Learning, Deep Learning, Streamlit Dashboard, Risk Prediction.

Modern financial services depend on automated underwriting and risk assessment to price premiums, extend credit, and catch fraud. However, traditional statistical models often struggle with today’s data which is increasingly high-dimensional, non-linear, and time-sensitive. Gradient Boosting Algorithms (GBAs), such as XGBoost, LightGBM, and CatBoost, have stepped in to fill this gap, offering the flexibility needed to process complex, diverse data sources.This study dives into how these frameworks can be optimized for financial risk. We move beyond basic algorithmic theory to address the "missing pieces" of implementation: feature representation, model interpretability, and the merging of historical records with real-time behavioral data. This reseacrh also introduce a hybrid modeling approach that syncs behavioral scoring with transactional patterns to sharpen predictive accuracy. Our testing on benchmark datasets shows that these optimized models don't just perform better they provide the stability and discrimination power necessary for high-stakes financial decisions. This study addresses the trade-offs between raw performance and the ethical necessity of fairness and transparency in automated systems.

Open Source Software (OSS) has become a very important and crucial infrastructure worldwide because of the value it provides. OSS typically depends on contributions from developers across diverse backgrounds and levels of experience. Making safe changes, such as fixing a bug or implementing a new feature, can be challenging, especially in object-oriented systems where components are interdependent. Static analysis and defect-prediction tools produce metrics (e.g., complexity, coupling) that flag potentially fault-prone components, but these signals are often hard for contributors new or unfamiliar with the codebase to interpret. Large Language Models (LLMs) have shown strong performance on software engineering tasks such as code summarization and documentation generation. Building on this progress, we investigate whether LLMs can translate faultprediction metrics into clear, human-readable risk explanations and actionable guidance to help OSS contributors plan and review code modifications. We outline explanation types that an LLM-generated assistant could provide (descriptive, contextual, and actionable explanations). We also outline our next steps to assess usefulness through a task-based study with OSS contributors, comparing metric-only baselines to LLM-generated explanations on decision quality, time-to-completion, and error rates.

The second leading cause of death for women worldwide remains breast cancer. In particular as they age, all women have a higher risk of acquiring breast cancer. Populations range in their incidence of various forms of breast cancer, and they also differ in their risks of developing breast cancer in general. Hence, it is crucial to determine the degree of the malignancy as soon as possible. Artificial intelligence may be able to classify breast cancer with competence levels on par with those of a doctor. This paper proposes the detection of breast cancer through a deep learning algorithm-based approach that involves segmentation, feature extraction, and grading. A soft computing technique is developed to categorize tumours as benign or malignant using breast histopathology images. To determine the exact grades of breast cancer, the initiation technique uses K-mean clustering-based segmentation, feature extraction, and classification. The segmented region yielded attributes such as correlation, homogeneity, contrast, LBPs (local binary patterns), and energy coefficients. In order to establish which network is better at recognizing the diagnosis of breast malignancy, the Artificial neural network classifier, deep learning architectures of AlexNet and GoogLeNet are contrasted. The AlexNet classification methodology produces accurate solutions than other methods by obtaining 100% accuracy. The proposed enhanced segmentation and classification techniques may help the automated diagnosis system to reduce the number of erroneous diagnoses and increase prediction performance. It could therefore aid clinicians in obtaining an additional opinion and in the early detection of diseases.

No abstract available

Peritumoral edema regions carry prognostic value in patients with high-grade glioma (HGG), the most invasive type of brain cancer. Recent findings have established the association of texture and shape features extracted from these regions with survival outcomes. However, no study has converged on a single feature that significantly correlates with survival outcomes. In this study, we develop an automated and interpretable brain tumor patient survival risk prediction model using radiomic features from the peritumoral region of HGG. First, the peritumoral edema regions are segmented from MRI scans imaged using multiple modalities (T1, T2, FLAIR, T1-contrast enhanced) and compiled into the BraTS-2020 dataset. Texture and shape features extracted from the segmented regions were analyzed to stratify patients based on a risk score. The proposed framework demonstrates the significance of a texture and shape feature to predict survival outcomes for a subset of 76 HGG patients with survival information. Moreover, we conduct univariate and multivariable analysis to further demonstrate the clinical utility of the extracted texture and shape features. The study provides evidence for the importance of texture and shape features extracted from peritumoral edema regions in predicting survival outcomes in HGG patients. It may facilitate personalized treatment and improve the prognostic accuracy of HGG patients in real-world clinical setting.

Software requirement specification (SRS) document is the most crucial document in software development process. All subsequent steps in software development are influenced by this document. However, issues in requirement, such as ambiguity or incomplete specification may lead to misinterpretation of requirements which consequently, influence the testing activities and higher the risk of time and cost overrun of the project. Finding defects in the initial development phase is crucial since the defect that found late is more expensive than if it was found early. This study describes an automated approach for detecting ambiguous software requirement specification. To this end, we propose the combination of text mining and machine learning. Since the dataset is derived from Malaysian industrial SRS documents, this study only focuses on the Malaysian context. We used text mining for feature extraction and for preparing the training set. Based on this training set, the method ‘learns’ to detect the ambiguous requirement specification. In this paper, we study a set of nine (9) classification algorithms from the machine learning community and evaluate which algorithm performs best to detect the ambiguous software requirement specification. Based on the experiment’s result, we develop a working prototype which later is used for our initial validation of our approach. The initial validation shows that the result produced by the classification model is reasonably acceptable. Even though this study is an experimental benchmark, we optimist that this approach may contributes to enhance the quality of SRS.

No abstract available

Multi-Criteria Decision-Making (MCDM) techniques play a critical role in solving complex decision problems involving multiple conflicting criteria across various domains. The Preference Ranking Organization Method for Enrichment Evaluations (PROMETHEE) is a widely adopted MCDM method known for its structured and transparent ranking process. However, conventional PROMETHEE implementations often rely on manually assigned criteria weights, introducing subjectivity and inconsistency in decision-making. This study proposes a novel hybrid approach that integrates Machine Learning (ML) techniques with PROMETHEE to enhance decision-making, particularly in blockchain security risk assessment. The proposed ML-PROMETHEE model leverages ML algorithms for feature selection, weight assignment, and preference function optimization, ensuring an automated and data-driven ranking process that minimizes human bias and enhances decision reliability. A real-world case study in blockchain security risk assessment is conducted to validate the proposed methodology. This study employs ML models such as Random Forest (RF) for feature selection and SHapley Additive exPlanations (SHAP) analysis to determine the most influential security risk factors. Additionally, Artificial Neural Networks (ANNs) and eXtreme Gradient Boosting (XGBoost) are used to optimize weight assignment, refining the decision-making process before integrating it into the PROMETHEE framework. Experimental results demonstrate that the ML-enhanced PROMETHEE model achieves a 17.3% improvement in decision accuracy, a 43.5% reduction in execution time, and a 40.2% reduction in subjective bias compared to traditional PROMETHEE-based decision-making. These findings highlight the potential of ML-enhanced MCDM models in strengthening the robustness, transparency, and scalability of decision frameworks, particularly in rapidly evolving domains such as blockchain security. Future research will explore the integration of deep learning and reinforcement learning techniques to further enhance automated decision-making frameworks.

This study examines the pricing dynamics of Non-Fungible Tokens (NFTs) in the secondary market using advanced machine-learning techniques. We construct a large dataset of Ethereum-based NFT transactions initially comprising over 500,000 raw blockchain observations spanning multiple NFT segments, including art, collectibles, gaming, metaverse, and utility assets, over the period from November 2018 to March 2023. Following data preprocessing, synchronization across data sources, and the construction of history-dependent features, the analysis focuses on a final analytical sample of approximately 70,000 transactions. To address the challenges of non-fungibility, thin trading, and high price dispersion, we develop an interpretable predictive framework that integrates domain-informed manual feature engineering, automated Deep Feature Synthesis, and dimensionality reduction via Principal Component Analysis. Three non-linear models—Random Forest, XGBoost, and a Multilayer Perceptron—are trained and evaluated using both random and time-aware validation strategies. The results indicate that XGBoost consistently achieves the highest predictive accuracy, both overall and across individual NFT segments, while historical transaction prices emerge as the dominant predictor of future prices. Segment-level analysis reveals substantial heterogeneity in predictability, with art and collectible NFTs exhibiting more stable pricing patterns than gaming and metaverse assets. Overall, the findings highlight strong path dependence and reputation-driven valuation in NFT markets and demonstrate that carefully designed machine-learning models can deliver high predictive performance without sacrificing economic interpretability.

. Online transaction grows in enormous rate because of the strength of usage by the user. User always use online mode to pay the amount to the respective merchant. Various method of payment is available in the market but credit card is so popular due to the pre credit is assigned to the customer by banker. Card user gets extra time for paying the payment which gives comfortable live to them. Security of the card suffers in various factors such as theft, fraud, illegal access, so it is protected by using modern algorithm with automated capability. Artificial Intelligent algorithms are applied to detect the fraud but that is not achieving enough accuracy. This type of problem is overcome by using location based risk identification model with multidimensional features for analysis. Three phases of processing is carried out namely feature management, risk management and Location awareness. The focus of the model is to protect the credit card frauds in multi level security by identifying the source and location of access. It achieves high level of security when compared to all exiting algorithms with reliable manner.

No abstract available

Amid the increasing complexity and dynamic nature of financial markets, accurately capturing market fluctuations and implementing effective risk monitoring remain critical challenges in financial regulation. Traditional differential equation models, while proficient in theoretical derivation and variable representation, face significant limitations in handling high-dimensional, complex data and nonlinear characteristics. Conversely, deep learning technologies, with their robust feature extraction and time-series modeling capabilities, present transformative opportunities for financial data analysis. However, the trade-off between high-precision modeling and interpretability creates notable challenges for single-method approaches. To address these limitations, this study proposes a dynamic optimization framework, CT-BCIR, which integrates traditional differential equations with deep learning methodologies. The framework employs convolutional neural networks (CNNs) to extract local temporal features and long short-term memory networks (LSTMs) to capture long-range dependencies.

Indoor-outdoor detection (IOD) has gained prominence recently, particularly in positioning technology, leveraging smartphone-embedded sensors. It is pivotal in pedestrian localization, activity recognition, transportation mode classification, and power management of Internet of Things (IoT) devices. While several approaches have been explored for IOD, including threshold-based methods and machine learning-based models, challenges remain in addressing these models’ temporal variations and computational complexities. Supervised learning approaches heavily rely on labeled datasets, which are costly and time-consuming to synthesize. We propose TabCLR, the first self-supervised learning (SSL) framework for IOD, to overcome these challenges. TabCLR utilizes contrastive learning representation tailored for tabular data classification using smartphone inertial sensors. It comprises data augmentation, a novel encoder network with self-attention, and an optimized contrastive loss function. Evaluation of TabCLR on multiple indoor-outdoor datasets demonstrates its superiority in both supervised and semi-supervised classification compared to existing methods. Notably, TabCLR outperforms SCARF by 6%-7%, indicating its effectiveness in capturing temporal feature representation patterns. Visualization analysis further illustrates TabCLR’s distinctive clustering of feature embeddings compared to SCARF. TabCLR represents a significant advancement in SSL methodologies for indoor-outdoor detection classification. Its robust performance showcases its potential to enhance accuracy in indoor-outdoor integrated GPS systems, addressing critical challenges in IOD classification.

Abstract Credit risk management is crucial for the credit loan decision-making process on P2P platforms, essential for mitigating credit defaults. The rapid decline of China’s P2P platform market highlights the urgent need for Internet financial institutions to enhance their credit risk management strategies. Previous studies have applied machine learning to assess credit risk; however, their effectiveness is often hampered by a lack of consideration for semantic information and individual correlation. In response, we propose an innovative approach, KG-GNN, which integrates knowledge graph (KG) and graph neural network (GNN). KG-GNN leverages KG to encapsulate semantic information within complex categorical features and explore potential relationships between borrowers. Utilizing GNN, our framework extracts representation features from the KG to build comprehensive and accurate credit risk models. Our findings indicate that KG-GNN not only can predict credit risk more accurately than conventional machine learning models but also improves the stream model performance by over 20% through KG and GNN-based data augmentation techniques. By integrating KG with GNN, our approach enriches the methodologies for credit risk management and can be adapted to other data mining challenges that require processing complex semantic and relational information, thereby enhancing model learning capabilities.

Assessment and management of credit risk at banks is a critical factor that ensures the stability and profitability of these institutions. Existing traditional statistical approaches that worked in the past are already proving to be incapable of coping with this new environment and the complexities intertwined within modern financial markets. Modern methodologies like probabilistic graphical models (PGMs) provide sophisticated methods for modeling these complex relationships, which integrate graph theory with probability theory. In this paper, we will explore Credit Risk and its main components, the mathematical foundation behind credit risk assessment, and the modeling techniques of these components. It compares traditional statistical models (eg, logistic regression and Monte Carlo simulations) with advanced Probabilistic Graphical models (PGMs). The paper highlights selecting the latter based on its capacity for more accurate representation of complex dependencies and uncertainties. PGMs that are covered here include Bayesian and Markov Networks, specifically for their structural representation of joint probability distributions, conditional independence as well as efficient inference. PGMs stand out for an improved model of non-linear interactions, and they allow the incorporation of uncertainty in a natural way, dynamic updating and systematic risk segmentation. This includes using Expectation-Maximization and Gradient-Based Optimization — bringing machine learning and modern computational methods to PGMs. It exemplifies the practical use of PGMs in credit risk management with examples ranging from default probability prediction to portfolio risk assessment and real-time risk monitoring. In conclusion, this paper points to the future promise of PGMs in credit risk management through continuing advancements in computation facilitated by embedding within an ML/AI framework. Reimagining This transformation will revolutionize how financial institutions measure and predict risks.

Various types of drug toxicity can halt the development of a drug. Because drugs are xenobiotics, they inherently have the potential to cause injury. Clarifying the mechanisms of toxicity to evaluate and manage drug safety during drug development is extremely important. However, toxicity mechanisms, especially hepatotoxic mechanisms, are very complex. The significant exposure of liver cells to drugs can cause dysfunction, cell injury, and organ failure in the liver. To clarify potential risks in drug safety management, it is necessary to systematize knowledge from a consistent viewpoint. In this study, we adopt an ontological approach. Ontology provides a controlled vocabulary for sharing and reusing of various data with a computer-friendly manner. We focus on toxic processes, especially hepatotoxic processes, and construct the toxic process ontology (TXPO). The TXPO systematizes knowledge concerning hepatotoxic courses with consistency and no ambiguity. In our application study, we developed a toxic process interpretable knowledge system (TOXPILOT) to bridge the gaps between basic science and medicine for drug safety management. Using semantic web technology, TOXPILOT supports the interpretation of toxicity mechanisms and provides visualizations of toxic courses with useful information based on ontology. Our system will contribute to various applications for drug safety evaluation and management.

: Household financial vulnerability reflects the likelihood that a household will fall into financial distress when facing adverse shocks, and it is a key micro-foundation of financial system stability. Using microdata from the China Household Finance Survey (CHFS), this study develops a machine-learning-based risk management framework to examine how liquidity constraints affect household financial vulnerability and to identify high-risk groups under heterogeneous socioeconomic conditions. Household financial vulnerability is operationalized as a binary outcome, denoted as the Financial Vulnerability Index (FVI), indicating whether liquid buffers are sufficient to cover unexpected expenditures. Liquidity constraints are measured through credit accessibility indicators, forming a binary Liquidity Constraint (LC) variable. The framework integrates (i) high-performance tabular prediction models, including gradient-boosted decision trees and neural tabular networks, to construct calibrated probability-of-vulnerability scores; (ii) explainability techniques, with Shapley Additive Explanations (SHAP) used to quantify global and local risk drivers; and (iii) causal machine learning methods, such as Double Machine Learning (DML) and generalized random forests, to estimate the heterogeneous causal effect of liquidity constraints on financial vulnerability across income groups, city tiers, and regions. To enhance model reliability for risk governance, probability calibration and distribution-free uncertainty quantification are implemented via conformal prediction. Empirical results indicate that liquidity constraints significantly increase the predicted and causally estimated risk of household financial vulnerability, with stronger effects concentrated among middle-to-lower income households and households located in lower-tier cities and economically stressed regions. The proposed framework provides an algorithmic basis for targeted inclusive-finance interventions and household risk mitigation policies.

Information Security Risk Management (ISRM) is an essential requirement for organizations seeking to ensure the governance and protection of their information assets. Ontology-based knowledge representation has emerged as a promising solution to address information security challenges, as it enables the formalization of concepts, relationships, and constraints within a given domain. This paper proposes an ontology-based framework aligned with the ISO/IEC 27002 standard. The approach consists of extracting relevant concepts from textual sources using UML modeling and TF-IDF filtering, and representing them in OWL using the Protégé environment. The resulting ontology formally captures key ISRM entities—including assets, threats, vulnerabilities, risks, controls, and monitoring mechanisms. The ontology was validated using the FACT++ reasoner to assess consistency and semantic completeness. The results show that the proposed model ensures traceability across ISO/IEC 27002 control families, supports governance alignment, and improves visibility across risk treatment processes.

IT project management faces critical challenges related to inaccurate resource allocation estimation and project risk assessment, which complicates decision-making and threatens project performance. Although machine learning techniques have been widely adopted in this domain, existing studies predominantly rely on single models or simple ensemble strategies, limiting their ability to capture heterogeneous interactions among organizational, technical, and risk-related factors. This study proposes a hybrid machine learning–based decision support framework that integrates feature-level representation learning and probabilistic decision fusion. Gradient Boosting is reconceptualized as a feature selection and nonlinear interaction modeling mechanism, while Artificial Neural Networks generate latent feature embeddings representing complex project characteristics. These representations are fused through a Naive Bayes classifier to produce calibrated probabilistic predictions, supported by a weighted fusion strategy with F1-score–based threshold optimization to improve stability under imbalanced risk conditions. Experimental evaluation is conducted using 5,997 synthetic IT project records from PT Anugerah Nusa Teknologi. Model performance is evaluated using accuracy, precision, recall, F1-score, and ROC-AUC metrics. Compared to standalone Gradient Boosting, Artificial Neural Network, and Naive Bayes models, the proposed hybrid framework consistently demonstrates superior predictive performance, achieving an accuracy of 0.85, an F1-score of 0.8485, and a ROC-AUC of 0.9050. Theoretically, this study contributes to project management research by demonstrating that IT project outcomes are more effectively modeled through multi-perspective learning rather than isolated predictors. Practically, the proposed framework provides actionable decision support to assist project managers in optimizing resource allocation and prioritizing risk mitigation under uncertainty.

No abstract available

No abstract available

The complexity of the business environment, which is shaped by dynamic economic, environmental, and social factors, makes it more important for companies to develop sustainable risk management practices. This research proposes a novel approach that combines traditional methods with modern machine learning techniques in response to the complex challenges faced by contemporary businesses in balancing risk and sustainability. Our study uses a public bank loan default dataset as a case study to address missing data systematically through robust imputation mechanisms and transform categorical variables using feature encoding. Spearman correlation analysis helps us understand complex variable relationships and guides subsequent feature selection. The decision tree classifier, a powerful machine learning algorithm known for its interpretability, is applied to identify key factors contributing to risk assessment. The hierarchical structure of the decision tree not only reveals important variables but also provides an explicit representation of the decision-making process. ROC curve analysis shows how well our predictive model can differentiate potential loan defaults.

The Life Detection Knowledge Base (LDKB; https://lifedetectionforum.com/ldkb) is a community-owned web resource that is designed to facilitate the infusion of astrobiology knowledge and expertise into the conceptualization and design of life detection missions. The aim of the LDKB is to gather and organize diverse knowledge from a range of fields into a common reference frame to support mission science risk assessment, specifically in terms of the potential for false positive and false negative results when pursuing a particular observation strategy. Within the LDKB, knowledge sourced from the primary scientific literature is organized according to (1) a taxonomic classification scheme in which potential biosignatures are defined at a uniform level of granularity that corresponds to observable physical or chemical quantities, qualities, or states; (2) a set of four standard assessment criteria, uniformly applied to each potential biosignature, that target the factors that contribute to false positive and false negative potential; and (3) a discourse format that utilizes customizable, user-defined “arguments” to represent the essential aspects of relevant scientific literature in terms of their specific bearing on one of the four assessment criteria, and thereby on false positive and false negative potential. By mapping available and newly emerging knowledge into this standardized framework, we can identify areas where the current state of knowledge supports a well-informed science risk assessment as well as critical knowledge gaps where focused research could help flesh out and mature promising life detection approaches.

Model‐based systems engineering is a powerful methodology to develop safety‐critical systems. The use of the system model as a single source of truth for risk and dependability analysis results in a consistent and complete assessment. Besides, representation and logging of the assessment within the model result in a complete and up‐to‐date single source of information that can be used during the device certification as well. This paper aims to provide a comprehensive risk management SysML profile that includes interconnected safety analysis [functional hazard assessment (FHA), fault tree, and failure mode and effect analysis (FTA, FMEA)], control measure, and evaluation model elements in compliance with the medical standards. Model‐based risk assessment of a point‐of‐care diagnostic device for sepsis has been shown as a case study to show the implementation of the profile. This device is a standalone unit and the test results obtained directly affect the patient. Therefore, both the top‐down (FHA and FTA) and bottom‐up (FMEA) safety assessment methods have been used. Another objective of the study is to define a systematic and holistic method to perform fault tree analysis, not only from the system architecture models but also from the functional, activity, and sequence diagrams of the system model.

Cardiovascular disease (CVD) remains one of the leading causes of global mortality and disability. Advances in computational modeling and artificial intelligence have enhanced CVD risk prediction by integrating multivariate clinical and biochemical features. Recent developments in deep learning, especially transformer-based models for tabular data, have demonstrated superior capabilities in capturing nonlinear and high-dimensional biomarker interactions. This study proposes a predictive framework that uses statistical and clinical biomarker data to assess CVD risk. Traditional machine learning models (logistic regression, Gaussian Naïve Bayes, linear discriminant analysis, AdaBoost, and XGBoost) were compared with deep learning models (gated recurrent unit [GRU] and long short-term memory [LSTM]) and transformer-based models—self-attention and intersample attention transformer (SAINT), feature tokenizer (FT), and tab transformer. Experiments were conducted to evaluate the impact of data augmentation, analyze learning behavior through loss and accuracy curves, and assess a fusion approach combining tab transformer with recurrent networks. Model performance was evaluated using accuracy and receiver operating characteristic analysis. Transformer-based models consistently outperformed conventional machine learning and deep learning methods. SAINT and FT achieved an area under the curve (AUC) of 0.8875 and 0.9489, respectively. The tab transformer demonstrated the highest performance with an AUC of 0.9728. The fusion of the tab transformer with GRU and LSTM further enhanced predictive precision, improving representation learning and generalization for CVD risk prediction. The proposed transformer-based framework offers a robust, scalable, and interpretable solution for accurate CVD risk assessment. Its superior predictive capability highlights the potential for integration into clinical decision-support systems for early diagnosis and patient management.

Multi-label medical image classification is challenging due to complex inter-label dependencies, data imbalance, and the need to integrate multiple data modalities. These challenges hinder the development of robust and interpretable diagnostic systems capable of leveraging diverse clinical information. We propose a cancer risk stratification framework that combines univariate thresholding with multivariate modeling using a hybrid parallel deep learning architecture, MedFusionNet. First, univariate thresholds are applied to identify the top-N discriminative features for each label. These selected features are then incorporated into MedFusionNet, which integrates Self-Attention Mechanisms, Dense Connections, and Feature Pyramid Networks (FPNs). The architecture is further extended for multi-modal learning by fusing image data with corresponding textual and clinical metadata. Self-Attention captures dependencies across image regions, labels, and modalities; Dense Connections enable efficient feature propagation; and FPNs support multi-scale representation and cross-modal fusion. Extensive evaluations on multiple datasets, including NIH ChestX-ray14 and a custom cervical cancer dataset, confirm that MedFusionNet consistently outperforms existing models. The framework delivers higher accuracy, improved robustness, and enhanced interpretability compared to traditional deep learning approaches. MedFusionNet provides an effective and scalable solution for multi-label medical image classification and cancer risk stratification. By integrating multi-modal information and advanced architectural components, it improves predictive performance while maintaining high interpretability, making it well-suited for real-world clinical applications. Medical images play an important role in helping doctors assess a person’s risk of developing cancer. However, these images can be challenging for computer systems to interpret, especially when several findings appear together, or when pieces of clinical information (such as patient history or handwritten notes) are stored separately. We developed a more reliable computational approach that first identifies the most important clinical features linked to cancer risk. These features are then analyzed using a model called MedFusionNet, which brings together information from medical images, clinical data, and text. This combined approach helps the system recognize patterns that might be overlooked when each type of information is considered on its own. When evaluated on large public datasets and a separate clinical dataset, MedFusionNet showed more accurate and consistent results than other commonly used techniques. These improvements may support earlier detection of cancers, reduce uncertainty in cancer diagnosis, and help clinicians make clearer and more informed decisions. Gorbachev et al. present a cancer risk stratification framework that combines univariate thresholding with multivariate modeling using a hybrid parallel deep learning architecture. They find that integration of multi-modal information and advanced architectural components improves predictive performance while maintaining high interpretability.

Cardiac magnetic resonance (CMR) imaging is the gold standard for non-invasive cardiac assessment, offering rich spatio-temporal views of the heart's anatomy and physiology. Patient-level health factors, such as demographics, metabolic, and lifestyle, are known to substantially influence cardiovascular health and disease risk, yet remain uncaptured by CMR alone. To holistically understand cardiac health and to enable the best possible interpretation of an individual's disease risk, CMR and patient-level factors must be jointly exploited within an integrated framework. Recent multi-modal approaches have begun to bridge this gap, yet they often rely on limited spatio-temporal data and focus on isolated clinical tasks, thereby hindering the development of a comprehensive representation for cardiac/health evaluation. To overcome these limitations, we introduce ViTa, a step toward foundation models that delivers a comprehensive representation of the heart and a precise interpretation of individual disease risk. Leveraging data from 42,000 UK Biobank participants, ViTa integrates 3D+T cine stacks from short-axis and long-axis views, enabling a complete capture of the cardiac cycle. These imaging data are then fused with detailed tabular patient-level factors, enabling context-aware insights. This multi-modal paradigm supports a wide spectrum of downstream tasks, including cardiac phenotype and physiological feature prediction, segmentation, and classification of cardiac/metabolic diseases within a single unified framework. By learning a shared latent representation that bridges rich imaging features and patient context, ViTa moves beyond traditional, task-specific models toward a universal, patient-specific understanding of cardiac health, highlighting its potential to advance clinical utility and scalability in cardiac analysis. 2.

The complexity and volume of financial data are also rising and require more powerful methods and more intelligent approaches to successful credit risk evaluation. The weakness of many traditional supervised learning algorithms is that they are sensitive to the availability of labelled data, which is scarce, skewed, and costly to acquire in the credit setting. The paradigm of self-supervised learning (SSL), which relies on supervision by the data in its most basic form, has become an attractive alternative option, particularly where labelled data is scarce. This article examines progressive, self-guided instructions for illustrating credit information and risk classification. This is done by creating a system that can create meaningful embeddings of the credit profile of customers using pretext tasks like masked feature prediction, contrastive learning, and autoencoding. Thereafter, the downstream credit risk prediction tasks are performed using such embeddings. We do substantial experimentation on real-world data sets of credit, and compare our models to classic supervised approaches. We have shown that self-supervised models can show a similar or even better performance in credit risk stratification, especially in the setting where there is a limited number of labelled data available. Moreover, we debate the interpretability, deterioration, and generalization abilities of SSL-based models in financial use cases. We also give an insight into how different Tasks implemented in SSL and architecture options affect the goodness of representations learned. The paper ends with a debate on how self-supervised learning will transform risk management in financial services by helping them create fairer, precise, and efficient credit rating models.