Transactions Analyzed
Highest ROC-AUC (Stacking Ensemble)
Fraud Cases Correctly Detected
Financial institutions face a critical paradox: the most accurate fraud detection models are also the most opaque,
making them legally undeployable under emerging global regulations. This project proposes a two-phase explainable
ensemble learning framework that achieves both goals simultaneously — strong fraud detection AND full regulatory
transparency documentation aligned to EU AI Act Articles 10 and 13.
SMOTE-balanced training on 284,807 real transactions
Dual SHAP + LIME explainability layers
EU AI Act compliance documentation generated automatically