End-to-End Python implementation of Kanziga et. al's (2026) credit-risk research method: monotonic binning, Weight-of-Evidence encoding, Information Value filtering, residual gradient boosting with hybrid-AUC early stopping, adaptive weighting via a 27-point CV search, Platt calibration, TreeSHAP, DeLong tests & stratified paired bootstrapping.
python machine-learning reproducible-research jupyter-notebook xgboost logistic-regression weight-of-evidence credit-scoring residual-learning gradient-boosting probability-calibration credit-risk explainable-ai east-africa financial-inclusion interpretable-machine-learning algorithmic-fairness shap credit-scorecard underbanked
-
Updated
Sep 4, 2026 - Jupyter Notebook