Date of Publication : May 2026

Abstract: Machine learning tools are becoming more common in microfinance institutions for automatically assessing credit risk. These models often perform better than traditional statistical methods in predicting creditworthiness. However, their black-box nature creates important issues around transparency, fairness, regulation, and the rights of borrowers. In countries like India, where more than 190 million people still do not have access to financial services, unclear credit rejection decisions can worsen existing inequalities and reduce trust in digital lending systems. To address this, this paper introduces a complete Explainable AI framework that combines XGBoost with SHAP. The framework is tested using the "Give Me Some Credit" dataset (150,000 real-world loan applications). The system achieves a classification accuracy of 91.2% and an AUC-ROC of 0.943, outperforming Logistic Regression, Decision Tree, Random Forest, and LightGBM. SHAP analysis shows that credit card usage, frequency of late payments, and applicant age are the most important factors. The system provides detailed, easy-to-understand reports for each applicant, helping loan officers explain their decisions and allowing borrowers to improve their credit standing.

Reference :