EXPLAINABLE AI MODELS FOR CREDIT RISK SCORING IN BANKING: BALANCING ACCURACY AND REGULATORY TRANSPARENCY
Keywords:
Credit Risk Scoring, Explainable AI, SHAP, Accuracy–Transparency Trade-off, Banking RegulationSynopsis
Credit risk scoring remains the foundation of banking decisions, yet the sector faces a persistent tension between predictive accuracy and regulatory transparency. While prior research has examined machine learning (ML) techniques for credit scoring, comparatively little attention has been paid to systematically quantifying the trade-off between performance and explainability under supervisory requirements. This study evaluates five models—Logistic Regression, Decision Tree, Random Forest, Gradient Boosting, and a Deep Neural Network—on the German Credit dataset, integrating Shapley Additive Explanations (SHAP) to provide post-hoc interpretability. Unlike existing works that focus narrowly on accuracy or transparency in isolation, our analysis demonstrates how SHAP can bridge ensemble model performance with regulatory acceptance under Basel III and GDPR. Results show that Gradient Boosting offers the highest predictive power (AUC = 0.84), while Logistic Regression remains most compliant due to transparency. SHAP analysis highlights checking account status, loan amount, and credit duration as consistent drivers of credit risk. The contribution of this study lies in positioning explainable AI not merely as a technical layer but as a compliance-oriented tool, enabling financial institutions to adopt high-performing ML models without compromising accountability.
References
[1] Zhumashev.A.B, & Kuatbayeva A. A. (2025). Integrating Explainable AI Techniques into Credit Scoring Models: Striking a Balance Between Accuracy and Transparency. Theoretical Hypotheses and Empirical Results, (10). Retrieved from https://ojs.scipub.de/index.php/THIR/article/view/6233
[2] Brahmaji, K.K.P. (2024). Explainable AI in data analytics: Enhancing transparency and trust in complex machine learning models. International Journal of Computer Engineering and Technology, 15(5), 1054–1061 doi: https://doi.org/10.5281/zenodo.14012791
[3] Bussmann, N., Giudici, P., Marinelli, D. et al. Explainable Machine Learning in Credit Risk Management. Comput Econ 57, 203–216 (2021). https://doi.org/10.1007/s10614-020-10042-0
[4] Omkar Reddy Polu. (2024). AI-Based Fake News Detection Using NLP. International Journal of Artificial Intelligence & Machine Learning, 3(2), 231–239. doi: https://doi.org/10.34218/IJAIML_03_02_019
[5] Schmitt, Marc and Cummins, Mark, Beyond Accuracy in Artificial Intelligence Based Credit Scoring Systems: Explainability and Sustainability in Decision Support (August 9, 2023). Available at SSRN: https://ssrn.com/abstract=4536400 or http://dx.doi.org/10.2139/ssrn.4536400
[6] Herma Ruibo Qian. (2025). AI-Embedded Defi Protocols for Real-Time Credit Scoring and Risk Transfer Leveraging Cross-Chain Oracles and Quantum-Resilient Cryptography in Autonomous Fintech Ecosystems. International Journal of Fintech (IJFT), 4(1), 1-7. doi: https://doi.org/10.5281/zenodo.15761275
[7] Sjur (2022) : Explainable AI for credit assessment in banks, Journal of Risk and Financial Management, ISSN 1911-8074, MDPI, Basel, Vol. 15, Iss. 12, pp. 1-23, https://doi.org/10.3390/jrfm15120556
[8] Gaddam, B. K. (2024). Explainable AI (XAI) for Transparent Financial Decision-Making: A Technical Framework. International Journal of Research in Computer Applications and Information Technology, 7(2), 1688–1700. doi: https://doi.org/10.5281/zenodo.14244591
[9] de Lange PE, Melsom B, Vennerød CB, Westgaard S. Explainable AI for Credit Assessment in Banks. Journal of Risk and Financial Management. 2022; 15(12):556. https://doi.org/10.3390/jrfm15120556
[10] Venkata Raj Kiran Kollimarla, The Role of Artificial Intelligence in Democratizing Finance and Transforming Market Participation, International Journal of Advanced Research in Engineering and Technology (IJARET), 15(3), 2024, pp. 128-142. doi: https://doi.org/10.17605/OSF.IO/VRSPW
Published
Series
License
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
