Fraud Detection using XGBoost: A Machine‑Learning Approach
In this project I built and deployed a production‑ready fraud‑detection model with XGBoost. Starting from a highly imbalanced card‑payments dataset, I engineered risk‑signal features, used stratified cross‑validation and Bayesian hyper‑parameter tuning, then calibrated the decision threshold against the business cost–benefit curve. The final model achieves high recall while keeping false positives below tolerance, and SHAP explainability is surfaced so analysts can understand why each transaction was flagged.
