CrediCardDefaultPredModel

Developed a credit risk classification model to predict customer defaults using behavioral and financial features. Achieved an F2 score of 0.90 with XGBoost. SHAP analysis highlighted the importance of delay and payment behavior. The model aids in early risk detection and informed lending. This project successfully developed a credit risk classification model to predict customer defaults using behavioral and financial features. Through comprehensive feature engineering—including delay metrics, payment regularity, and interaction terms—the model captured critical patterns in customer behavior. After benchmarking multiple algorithms, XGBoost emerged as the most effective, achieving a high F2 score of 0.9017 and demonstrating superior recall for identifying defaulters. SHAP analysis confirmed the dominance of delay- and behavior-related features over raw financial indicators, validating the importance of behavior-based credit scoring. The final model was deployed on unseen data with an optimized threshold to maximize recall while maintaining precision. This pipeline demonstrates the practical application of machine learning for early risk detection in credit portfolios, aiding financial institutions in making informed lending decisions and minimizing potential losses. …