Open research questions in Financial Distress and Bankruptcy Prediction

What the literature leaves open

• for seeking to harness the inclusion potential of ML-based credit scoring while managing its risks, organized around the principles of fairness by design, transparency by requirement, and inclusion by measurement (Sabeena et al., 2025; Nwaimo 37 lending fairness pre-deployment et al., 2024). For federal regulators, the most urgent priority is the development of clear regulatory guidance on the use of alternative data in credit scoring under ECOA and FCRA frameworks, resolving the current ambiguity about which alternative data sources are permissible and what fairness standards apply to ML-based credit systems guidance that should be developed through a transparent notice-andcomment process involving financial institutions, civil rights organizations, consumer advocates, and technical experts (Johnson, 2019; Bartlett et al., 2022). The CFPB should testing establish mandatory requirements for ML credit scoring models used in consumer lending, analogous to the clinical trial requirements in drug approval, requiring prospective demonstration that a model does not produce disparate impacts across protected classes before it is deployed at scale in the consumer credit market (Hurlin et al., 2022). The Federal Reserve and OCC should the Community Reinvestment Act examination update framework to to credit alternative data-based underserved populations as qualifying CRA activity, creating positive regulatory incentives for financial institutions to invest in ML-based financial inclusion programs alongside their traditional branch-based community development activities (Loufield et al., 2018; Kshetri, 2021). For financial institutions, the evidence strongly supports investment in hybrid human-AI credit decision systems that use ML models to generate inclusion recommendations for thin-file applicants while retaining human review for edge cases and building institutional knowledge about ML-based credit performance over time an approach that captures the accuracy benefits of ML while managing regulatory risk and building the organizational learning capacity needed for responsible scaling (Fügener et al., 2022; Liberti & Petersen, 2019). Fintech companies operating ML credit platforms should adopt the Alternative Data Standards Framework proposed by the Center for Financial Inclusion (Loufield et al., 2018), which provides a structured approach to alternative data sourcing, quality assessment, and bias testing, and should publish annual fairness reports that allow external evaluation of their models' disparate impact across protected classes creating market accountability that complements but does not substitute for regulatory oversight.

  Predictive Analytics for Credit Accessibility: A Machine Learning Approach to Expanding Financial Inclusion in Underserved U.S. Communities · 2026 · DOI

• CreditR1 delivers calibrated PDs with evidence-grounded reasoning that supports internal model validation and human review; transferability beyond the Chinese A-share market remains an open empirical question.

  CreditR1: Calibration-Aware Reinforcement Learning for Interpretable Corporate Credit Risk Assessment with Large Language Models · 2026 · DOI

• Science 13. Khanom, F., Biswas, S., Uddin, M. S., & Mostafiz, R. (2024). XEMLPD: an explainable ensemble machine learning approach for Parkinson disease optimized diagnosis features.

  Explainable AI for Investor Suitability Assessment: Integrating Behavioral and Financial Signals in Institutional Onboarding · 2026 · DOI

• Originality/value This study challenges the “one-size-fits-all” approach to local government savings by examining predictors of deficits during the dot-com and Great Recessions and finding that socio-economic, fiscal, and financial health indicators are insufficient to predict the presence or magnitude of deficits during recessions.

  Machine learning to predict local government deficits during economic downturns · 2026 · DOI

• This paper presented a hybrid framework that integrates Generative Adversarial Net- works (GANs), Recursive Feature Elimination (RFE), and a Random Forest classifier for corporate bankruptcy prediction under extreme class imbalance. The proposed meth- odology effectively addresses two critical challenges in financial distress prediction: the scarcity of minority-class samples and the high dimensionality of financial features. By leveraging GAN-based data augmentation, the framework generates realistic synthetic minority-class samples, overcoming the limitations of traditional oversampling tech- niques in capturing complex data distributions. The proposed framework was evaluated on the Taiwan bankruptcy prediction dataset, which contains 6819 firms with only 2.6% bankrupt cases and 95 financial attributes. Experimental results demonstrate that the integration of GAN-based augmenta- tion, RFE-based feature selection, and threshold-optimized classification significantly improves minority-class detection performance. In particular, the proposed method increased recall from 0.20 (baseline Random Forest) to 0.386, while achieving a preci- sion of 0.63 and an F1-score of 0.479. Additionally, evaluation using imbalance-aware metrics such as ROC–AUC, PR–AUC, G-Mean, and Matthews Correlation Coefficient confirms the robustness of the proposed approach. Comparative analysis with baseline models and resampling techniques, including SMOTE and ADASYN, further validates the effectiveness of the proposed hybrid framework. The results also highlight the importance of a recall-oriented decision strategy in bank- ruptcy prediction, where minimizing false negatives is critical for financial risk manage- ment. The ablation analysis demonstrates that the combined integration of GAN-based data augmentation, feature selection, and threshold optimization yields superior perfor- mance compared to individual components, indicating a strong synergistic effect. Despite the strong performance achieved by the proposed framework, certain limita- tions should be acknowledged. The current evaluation is conducted on a single bench- mark dataset, which, although widely used and representative, may not fully capture the diversity of financial environments across different regions and industries. Future work will extend the validation of the proposed approach to additional real-world bankruptcy datasets to further assess its generalizability and robustness. In addition, the exploration of advanced generative models such as Conditional GANs, Wasserstein GANs, and CTGAN may further improve the quality and stability of synthetic data generation. Incorporating model interpretability techniques, such as SHAP, will also provide deeper insights into feature importance and enhance the trans- parency of the decision-making process, which is critical in financial risk assessment applications.

  Corporate bankruptcy prediction using generative adversarial network-based data balancing, recursive feature elimination, and random forest classification · 2026 · DOI

• In 2020-2021, firms showed good results due to the low interest rates, but when the rate increased in 2022-2023, they became highly risky; therefore, all benchmark models excluding HPAF and Merton model did not account for this scenario.

  Predictive Analytics in Corporate Financial Risk Assessment and Management · 2026 · DOI

• https://doi.org/10.1177/22785337221098287. Research, 401–424. 11(3), and Bayakhmetova, A., Rudenko, L., Krylova, L., Suleimenova, B., Niyazbekova, S., & Nurpeisova, A. (2025). Artificial intelligence in financial behavior: Bibliometric ideas and new opportunities. Journal of Risk and Financial Management, 18(3), 159. https://doi.org/10.3390/jrfm18030159. Bermudez Vera, I. M., Mosquera Restrepo, J., & Manotas-Duque, D. F. (2025). Data mining for the adjustment of credit scoring models in solidarity economy entiimbalances. Risks, 13(2), 20. ties: A methodology for addressing class https://doi.org/10.3390/risks13020020. 33 Equilibrium. Quarterly Journal of Economics and Economic Policy, 21(1), 13–39 Bland, E., Changchit, C., Changchit, C., Cutshall, R., & Pham, L. (2024). Investigating the components of perceived risk factors affecting mobile payment adopof tion. 216. https://doi.org/10.3390/jrfm17060216.

  Beyond the score: AI-driven behavioural analytics in credit risk and consumer finance · 2026 · DOI

• There are many limitations of the present study that constitute significant directions for future research. To begin with, although the present study attempts to examine the internal organizational components of the Open Budget Index to explore struc- tural relations between them, it is limited in its ability to consider other socio-eco- nomic and political variables. In this connection, future studies will need to include external control variables, such as GDP per capita, V-Dem Democracy Index, and other measures related to institutional quality throughout history. The second shortcoming of the present study lies in its cross-sectional nature, which allows for a descriptive overview of the phenomenon at hand but lacks the dimensionality for an analysis of dynamics and causal “turning points” where the country moves from one cluster of transparencies to another over time. In addition, although SHAP and LIME are great tools for interpreting local and global feature importance, they cannot demonstrate causation from an econometric standpoint. Future research might leverage instrumental variable approaches or the difference- in-differences methodology to test whether a certain reform in audit independence causes better OBI results. Lastly, cultural and informal institutions may play an important yet understudied role in explaining variations in budget transparency lev- els. Including sentiment analysis from social media in the existing machine learning model would allow one to capture behavioral incentives of fiscal transparency in a more thorough manner. Author contributions All aspects of the study, including conception, design, data collection, analysis, and manuscript preparation, were carried out solely by S. Ç., S. Y. and G.Ç. All authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by S. Ç., S. Y. and G.Ç. The first draft of the manuscript was written by S. Ç., S. Y. and G.Ç. All authors com- mented on previous versions of the manuscript. All authors read and approved of the final manuscript. Funding The authors declare that no funds, grants, or other support were received during the preparation of this manuscript. Unveiling structural patterns of global budget transparency… 24 Page 36 of 39 S. Çelik et al. Data availability The analyses in this study are based on publicly available Open Budget Index (OBI) data published by the International Budget Partnership. All data used in the manuscript are fully acces- sible through the International Budget Partnership’s online repository (https:// inter natio nalbu dget. org/ open- budget- survey/ count ry- resul ts). No proprietary or confidential data were used or generated. The authors did not produce any new datasets.

  Unveiling structural patterns of global budget transparency using advanced machine learning and explainable artificial intelligence frameworks · 2026 · DOI

• This finding also resonates with Simon's (1957) theory of bounded rationality, which posits that decision makers are constrained by limited information and cognitive capacity.

  Artificial intelligence and financial decision making in Indian banks: Adoption, effectiveness, and challenges · 2026 · DOI

• Advanced Machine Learning Models • The next versions of the application will use advanced machine learning and deep learning algorithms like XGBoost, Gradient Boosting, and Neural Networks to enhance the accuracy of predictions. • The advanced machine learning models will also assist in analyzing and identifying complex patterns in the applicant’s data to increase the reliability of the recommendation. 7.2 Real-Time Credit Score Integration • The loan sales prediction application will have the ability to integrate with real-time credit score providers that allow the loan companies to automatically obtain up-to-date credit information for their loan applications. • The integration of real-time credit score providers in the loan application also enhances the accuracy of assessing a customer’s eligibility for a loan while reducing the need for manual verification. 7.3 AI-Based Risk Analysis • The loan application prediction application will also include the use of Artificial Intelligence to conduct in-depth risk evaluations based on the behavior of the applicant, as well as the applicant’s transactional history and financial patterns. • The enhancement of using AI for risk analysis will allow lending institutions to identify potential loan defaults more accurately. 7.4 Mobile Application Development • A dedicated mobile application for the loan sales prediction application can be developed for both Android and iOS devices to make the loan sales prediction application accessible to users from their mobile devices. • Users will now have the ability to submit a loan request and view the loan prediction results from their mobile devices. 7.5 Cloud-Based Deployment • The future of the loan sales prediction application will include the ability to deploy the application on cloud computing platforms, which will increase the scalability, availability, and performance of the application. • The loan application will be equipped with cloud computing capabilities to allow users to gain access to a large number of loan applications. 8. CONCLUSION The Loan Approval Prediction System will provide a cost-effective, precise, and intelligent way to determine how likely an application will be approved. The Loan Approval Prediction System applies machine learning techniques to evaluate an applicant’s characteristics such as income, employment status, education, previous credit history, and the total amount of the loan requested to help determine if the potential borrower meets the criteria for being approved for a loan. By automating the evaluation process of loans, the Loan Approval Prediction System requires less manual effort to process loans; therefore, there will be less time spent approving loans, and decisions can be supported by data analysis within banks and other financial institutions.

  Loan Approval Prediction System · 2026 · DOI

• Enhances model adaptability and improves hyperparameter optimization in non-convex spaces Well-rounded, interpretable, efficient, and resilient credit risk prediction framework 2.5. Evaluation metrics Accuracy is the fraction of correctly classified instances in all the evaluated samples. It indicates the overall performance of the model across all classes. However, despite its popularity, it can be misleading when applied to an unbalanced dataset. Accuracy = TP + TN TP + TN + FP + FN (1) Precision is the measure of the positive instances that were expected and actually turned out to be positive. To be specifically very accurate, only in the case when false positives are the ones that bring the heavy consequences, precision is very critical. Such a model of high precision can yield reliable, low-error outcomes. Precision = TP TP + FP (2) Recall measures the ability of a model to correctly identify all real positive cases. This parameter is very important in situations where missing positive instances could lead to serious consequences. A high recall means the model is effectively reducing the number of false negatives. Recall = TP TP + FN (3) ARTICLE IN PRESS ARTICLE IN PRESS ACCEPTED MANUSCRIPT ARTICLE IN PRESS ARTICLE IN PRESS Since the F1 Score is the harmonic mean of precision and recall, it gives a just assessment to both metrics. F1 score is the metric of choice in case of imbalanced class distributions or if there is a need to find a balance between precision and recall. Hence, it is a good indicator of the model's ability to detect true positives while maintaining a low number of false negatives. F1 = 2. Precision.

  Transforming credit risk evaluation in digital lending from black box models to transparent decisions · 2026 · DOI

• considerably. Random Forest (Breiman, 2001), an ensemble of decorrelated decision trees trained via bagging, demonstrated superior generalisation and robustness. Gradient Boosting Machines (Friedman, 2001), and their optimised types and XGBoost LightGBM (Ke et al., 2017), further advanced predictive performance and became the primary and principal paradigm scoring competitions and sector deployments.

  Towards Interpretable Credit Risk Assessment: A Comparative Study of XAI Techniques with Regulatory Compliance · 2026 · DOI

• explore counterfactual generation algorithms, GPU- accelerated DiCE implementations, or pre- computed counterfactual libraries indexed by applicant profile clusters, minimizing generation time to sub-second latency. Longitudinal explanation stability. Evaluating SHAP and LIME explanation consistency across numerous and manifold model retraining cycles would provide critical and vital evidence on the operational dependability of XAI systems in production, where idea drift consistently requires model revisions. Multi-modal credit data. This study concentrated exclusively on ordered and methodical tabular to the XAI data. Extending unstructured inputs; bank statement writing, transaction sequences, and social network signals; integration of NLP-based XAI would need procedures such as attention visualisation, SHAP for justification extraction, representing a notable and practically relevant inquiry frontier. User studies on explanation grasp. Empirical studies measuring how correctly loan officers and transformers, and framework applicants understand and act on SHAP and DiCE outputs; and whether their grasp leads to better decisions; would provide the human-centred validation this study technical results need. Quantum-enhanced credit scoring. Emerging work on Quantum Machine Learning (QML) for classification assignments increases the query of whether quantum-native models will need completely new XAI paradigms, as existing strategies such as SHAP and LIME are designed for classical architectures. This intersection of QML and XAI denotes a long-horizon but theoretically rich study direction. 6.8 Final Reflection and chance, economic responsibly. As The deployment of machine learning in credit risk assessment is no longer a query of whether but of algorithms make how consequential influencing financial decisions millions of individuals ‘admittance to housing, the education, responsibility to make these decisions clear, fair, and contestable is both a legal imperative and an ethical one. This study has demonstrated that Explainable AI is not merely a regulatory adherence mechanism but a truly valuable instrument for developing better, fairer, and more reliable credit scoring systems. SHAP capability to identify age-based discrimination that might otherwise persist hidden within a high-performing model illustrates that explainability and fairness are profoundly intertwined; you cannot have one without the other in high-stakes AI. The path forward for the financial sector is not to select between accuracy and transparency, but to embrace XAI as the system that makes both concurrently feasible. The four-tier deployment framework proposed in this study provides one in specific step along empirical evidence, aligned with regulatory necessities, and designed for the operational realities of modern credit organizations.

  Towards Interpretable Credit Risk Assessment: A Comparative Study of XAI Techniques with Regulatory Compliance · 2026 · DOI

• of traditional methods. The financial data-mining method proposed by Li et al. [19] employs fuzzy clustering to enhance risk assessment and early warning. This method groups financial data with ill-defined boundaries and extracts relevant patterns for predictive analysis via fuzzy clustering. This technique, which accounts for the imprecision and ambiguity of financial facts, overcomes the limitations of crisp clustering. The method employs multiple financial indicators, some imprecise, to enhance the reliability and precision of financial risk assessment. Liu et al. [20] studied information disclosure and financial data mining using fuzzy logic. This ARTICLE IN PRESS ARTICLE IN PRESS research simplifies the analysis of large financial datasets with missing or ambiguous data by employing fuzzy set theory. The algorithms improve data handling, allowing for more precise data extraction for regulatory compliance, financial reporting transparency, and risk management. The research shows that fuzzy logic can reduce ambiguity in financial data and provide deeper insights into financial disclosures. By combining qualitative expert opinions with quantitative financial measures, the FAHP method takes a revolutionary approach. The combination generates fuzzy weights for real-time audit decision-making. There are few published solutions that prioritize risk-based audit prioritizing and optimize resources in a changing financial environment. This is true even when audit systems use AI and process mining to examine transactions and investigate fraud. The paper proposes using FAHP to prioritize audit jobs by fuzzy-ranked financial concerns boosts audit efficiency and accuracy. It is a major improvement over prior methodologies, which often used predefined criteria or ignored financial audits' dynamic nature. The FAHP-based framework is being promoted as a new auditing tool that integrates financial experts' perspectives with real-time, risk-based financial- sector data. The method improves audit accuracy, resource distribution, and adaptability. The framework challenges traditional and AI-based financial audits with an efficient, interpretable, and adaptable alternative. The novel aspect of the proposed method is its adaptability across applications; it integrates quantitative financial data with qualitative expert opinion within a framework that can be customized to each case. Furthermore, it employs FAHP (Fuzzy Analytic Hierarchy Process) to rank audits in real time based on the risk they represent. This approach uses fuzzy logic to convert financial ratios and expert opinions into fuzzy weights. This is markedly different from the usual auditing solutions that rely on preset criteria or thresholds, or on AI-based models that require frequent retraining. This enables the system to respond promptly to changing financial conditions and emerging threats. Auditors will be able to trace each decision through to the fuzzy logic computations, since the system's decision-making process is interpretable and transparent. The approach provides a more adaptable, interpretable, and resource-optimized solution for current financial audits by allowing for flexible audit planning and focusing resources on the most severe financial challenges. This significantly improves the efficacy, precision, and efficiency of audits.

  Fuzzy algorithm-driven financial statement analysis and intelligent audit system design · 2026 · DOI

• In the future, it would be interesting to integrate real-time financial data streams and anomaly- detection mechanisms into the FAHP model to enhance dynamic responsiveness. Another interest is in extending the FAHP model to allow for adaptive learning, comprising a fuzzy weight based on audit results. It will pursue external validation across sectors and regulatory compliance gaps to verify scalability, robustness, and general applicability across different financial and operational environments.

  Fuzzy algorithm-driven financial statement analysis and intelligent audit system design · 2026 · DOI

• There is a lot of momentum in combining AI and ML with financial services. Changes are coming quickly with the potential for substantial impact on the industry. In this section, predictions about future transitions of AI/ML in financial services are explored, opportunities for future research and innovation are identified, and suggestions for stakeholders to prepare for future changes are offered. International Journal of Computational Intelligence Systemshttps://doi.org/10.1007/s44196-025-01110-01 3 Page 17 of 21 181 17 The coming few years are likely to witness a significant speedup in the adoption of AI and ML technology across financial institutions. A major forecast is that AI will increasingly become integrated into financial risk management. According to Khanday et al. (2025), the potential of AI to transform risk management is its capacity to analyze market conditions in real-time, thus allowing financial institutions to anticipate and counteract incoming threats in a timely manner. The increasing maturity of predictive analytics will enable banks to make data-driven, informed decisions that improve their risk avoidance measures and operations efficiency. Also, customer service personalization is expected to become a signature feature of financial services with the use of AI. Research (Aithal and Prabhu 2025), point out that banks will use sophisticated ML models to scan through tremendous amounts of customer data to enable them to personalize products and services based on individual tastes and actions. This move towards hyper-personalization will be expected to vastly enhance customer satisfaction and loyalty since clients are being offered products that suit their specific requirements. With continued advances in AI, the banking industry will also witness a departure from universal solutions towards more personalized experiences. In addition, AI capabilities will be used more widely in regulatory compliance. Because complex financial regulation will require novel means of meeting compliance requirements, AI can automate many aspects of compliance monitoring to help adhere to legal obligations more efficiently and effectively. The existing research (Singh et al. 2023), explained that this will also allow institutions to traverse the challenging regulatory environment while decreasing the risks of non-compliance risk. In the context of AI/ML in financial services, there are several important areas to explore for future research and development. One key area is the ethical implications of AI, and in particular the issue of bias in algorithmic decision making. In our view, it is vital to address the ethical issues associated with AI usage, especially in terms of bias in fostering fair, transparent and accountable AI applications.

  Operational Excellence Through AI and ML in Financial Services: A Comprehensive Review of Applications, Challenges, and Future Directions · 2026 · DOI

• The evaluation period for fraud detection is 6 months and for credit risk is 12 months, but the paper does not discuss performance stability beyond these windows or how concept drift and temporal changes affect model reliability.

  Leveraging Machine Learning for Real-Time Fraud Detection and Risk Assessment in Modern Fintech Platforms Aayush Bharat Mandavia and Anurag Shrivastava · 2026 · DOI

• Model performance monitoring is mentioned to track AUC, precision, recall, and false positive rate with automated alerts for performance degradation, but the specific degradation thresholds and remediation procedures are not detailed.

  Leveraging Machine Learning for Real-Time Fraud Detection and Risk Assessment in Modern Fintech Platforms Aayush Bharat Mandavia and Anurag Shrivastava · 2026 · DOI

• The NLP-based regulatory monitoring system incorrectly classified 5.8% of documents by relevance and 10.9% by affected business area, but the paper does not discuss approaches to improve accuracy or address misclassification risks in compliance applications.

  Leveraging Machine Learning for Real-Time Fraud Detection and Risk Assessment in Modern Fintech Platforms Aayush Bharat Mandavia and Anurag Shrivastava · 2026 · DOI

• The thin-file borrower segment achieved a default rate of 6.2% compared to portfolio average of 4.8%, representing a commercially acceptable risk premium, but the paper does not address long-term performance monitoring or strategies to reduce this higher default rate over time.

  Leveraging Machine Learning for Real-Time Fraud Detection and Risk Assessment in Modern Fintech Platforms Aayush Bharat Mandavia and Anurag Shrivastava · 2026 · DOI

• The credit risk model showed only modest improvement over traditional bureau-only models for prime borrowers (AUC improvement of 0.03), but the paper does not explain why alternative data provides limited benefit for this segment or how to improve performance in this category.

  Leveraging Machine Learning for Real-Time Fraud Detection and Risk Assessment in Modern Fintech Platforms Aayush Bharat Mandavia and Anurag Shrivastava · 2026 · DOI

• Future research should focus on piloting the system across different financial institutions and evaluating its adaptability across diverse markets, its ability to manage dynamic data and ethical considerations such as fairness and the explainability of Generative AI (GAI).

  Transforming financial decision-making with hybrid artificial intelligence (AI) · 2025 · DOI

• Whether configuration tables pre- dict more accurately than the nu- merical score tables remains to be seen; if they are even close in relative accuracy, I would prefer the con- figuration tables for their communica- tion potential as indicated above. com at UCSF LIBRARY & CKM on March 13, 2015 The major limitation of the pre- dictive information in Tables 3 and 4 is that the information is based on too few cases and on releasees of only one year. They take advantage of cur- vilinear relationships of a predictor with postrelease behavior and, on some predictors, of unique interrela- tionships limited to a few categories.

  Prediction Tables as Accounting Devices for Judges and Parole Boards · 1962 · DOI

• Future research should examine temporal modelling and benchmarking against XGBoost, LightGBM, CatBoost and deep-learning approaches.

  From prediction to decision support: explainable machine learning for schedule-delay risk in complex infrastructure projects · 2026 · DOI

• The paper evaluates systems on production data from a mid-size digital payment platform and a fintech lending platform, but does not discuss generalizability across different platform types, geographic regions, or regulatory jurisdictions.

  Leveraging Machine Learning for Real-Time Fraud Detection and Risk Assessment in Modern Fintech Platforms Aayush Bharat Mandavia and Anurag Shrivastava · 2026 · DOI

Most-cited papers in Financial Distress and Bankruptcy Prediction

• Predictably Unequal? The Effects of Machine Learning on Credit Markets · The Journal of Finance · 2021 · 430 citations
• Bankruptcy prediction for SMEs using transactional data and two-stage multiobjective feature selection · Decision Support Systems · 2020 · 321 citations
• Predicting bank insolvencies using machine learning techniques · International Journal of Forecasting · 2020 · 141 citations
• Rethinking SME default prediction: a systematic literature review and future perspectives · Scientometrics · 2021 · 130 citations
• Forecasting credit ratings of decarbonized firms: Comparative assessment of machine learning models · Technological Forecasting and Social Change · 2021 · 122 citations
• A study on credit scoring modeling with different feature selection and machine learning approaches · Technology in Society · 2020 · 120 citations
• Three-stage reject inference learning framework for credit scoring using unsupervised transfer learning and three-way decision theory · Decision Support Systems · 2020 · 111 citations
• Financial distress prediction using integrated Z-score and multilayer perceptron neural networks · Decision Support Systems · 2022 · 105 citations
• Credit growth, the yield curve and financial crisis prediction: Evidence from a machine learning approach · Journal of International Economics · 2023 · 75 citations
• Incorporating textual and management factors into financial distress prediction: A comparative study of machine learning methods · Journal of Forecasting · 2020 · 72 citations

Most recent work

• FedQuAD: Fast-Converging Curvature-Aware Federated Learning for Credit Default Prediction from Private Accounting Data · Mathematics · 2026
• Uncovering cross-organizational risk patterns: a machine learning approach to predicting financial fraud via chain leader attributes · Review of Managerial Science · 2026
• Enhanced predictive modeling for financial risk assessment using hybrid AI (ML & DL) on structured and unstructured data · Discover Artificial Intelligence · 2026
• A spatio-temporal machine learning model for mortgage credit risk: Default probabilities and loan portfolios · European Journal of Operational Research · 2026
• Credit Scoring Prediction for Small and Medium-Sized Enterprises Based on Alternative Data and Gradient Boosting Algorithms · Information Resources Management Journal · 2026
• Leveraging Machine Learning for Real-Time Fraud Detection and Risk Assessment in Modern Fintech Platforms Aayush Bharat Mandavia and Anurag Shrivastava · International Journal of Advanced Research in Science Communication and Technology · 2026
• AI-Driven Bankruptcy Prediction in Manufacturing SMEs: Comparing Machine Learning Techniques with Logistic Regression · Administrative Sciences · 2026
• Operational Excellence Through AI and ML in Financial Services: A Comprehensive Review of Applications, Challenges, and Future Directions · International Journal of Computational Intelligence Systems · 2026
• Impact of AI-Based Credit Scoring on Loan Risk Management in Indian Banks · Zenodo (CERN European Organization for Nuclear Research) · 2026
• AI in Credit Risk Assessment: Creating Customer Awareness about CIBIL Score · International Research Journal of Modernization in Engineering Technology & Science · 2026

Related topics in Business, Management and Accounting