Interpretability of Machine Learning Models

Machine learning models often achieve better predictive performance in credit risk management than classical regression models — but frequently remain non-transparent. The technical article provides a structured overview of current methods for the explainability of ML models (Explainable AI), including SHAP values, LIME and model-specific approaches.

The methods presented are evaluated using concrete application examples from credit risk. Particular attention is paid to the question of which methods are suitable for regulated banking operations and what requirements arise from the EU AI Act and ECB guidance.