Credit risk remains the single largest source of losses for the Indian banking system, and the persistence of non-performing assets (NPAs) has renewed interest in data-driven approaches to loandefault prediction. This study develops and evaluates a predictive-analytics framework for credit-risk assessment using a dataset of 1,200 retail and small-business loan accounts drawn from Indian commercial banks. We examine the borrower, loan, and macro-level factors most strongly associated with default and benchmark four classification models: logistic regression, decision tree, random forest, and gradient boosting (XGBoost). Features including credit (CIBIL) score, debt-to-income ratio, loan-to-value ratio, employment stability, and prior delinquency were engineered and evaluated. The gradient-boosting model achieved the strongest discrimination (AUC = 0.91, accuracy = 87.4%), outperforming logistic regression (AUC = 0.82). Debt-to-income ratio, CIBIL score, and prior delinquency emerged as the most influential predictors. The findings suggest that ensemble machine-learning models can materially improve earlywarning systems for Indian banks while remaining interpretable through feature-importance analysis. The study offers practical implications for credit-scoring policy, provisioning, and regulatoy compliance under the RBI framework.

Pratik G. Khandekar, Dr. Ujwala Narkhede, "Predictive Analytics for Credit Risk Assessment in Indian Banks: An Empirical Study of Factors Associated with Loan Defaults" International Journal of Scientific Research and Engineering Development, V9(5): Page(1123-1126) September - October 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.