A Multi-Model Comparative Framework for Credit Risk Prediction: Statistical vs Machine Learning Approaches

Abstract

There are two domains in credit risk- retail credit risk & corporate credit risk. Traditional statistical models are used for creating default risk prediction. This study presents dual track Machine Learning pipeline for credit risk prediction compared to old & traditional statistical models. Logistic regression, Light GBM, XGBoost & Random Forest are used by this model to predict credit risk. This research reveals that ML models meaningfully out pace old-style statistical methods in both credit risk domains. For that, this study incorporates featured engineering, data acquisition & model interpretability for multistage analysis. This research study further argues interpretability through SHAP by ensuring regulatory compliance & transparency in model building.