YAI Connect: Development of an Early Warning System (EWS) for Banking Credit Risk

This master’s thesis presents the development of an Early Warning System (EWS) for banking credit risk, focused on the one-year Probability of Default (PD) within the Basel III framework. Using a synthetic retail credit portfolio and a Python-based implementation, several machine learning models are compared to assess their predictive power and robustness with respect to the nature and complexity of the problem.The analysis highlights the strong performance of boosting techniques in capturing non-linear relationships and complex risk patterns. Beyond model accuracy, special attention is given to variable behavior across risk ranges, model interpretability, and threshold calibration, all of which are critical for practical risk management applications.The results confirm that advanced AI-driven EWS can complement traditional credit risk frameworks by providing timely, interpretable, and operationally relevant signals. The study emphasizes a prudent integration of machine learning into regulatory-aligned decision processes, not only from a quantitative perspective but also from a qualitative one, as the proposed EWS incorporates both approaches, supporting earlier intervention and improved financial stability.