An Explainable Machine Learning Model for Credit Risk Prediction: Evidence from Commercial Banks in Bangladesh

With the increasing complexity of financial data, accurate credit risk prediction has become essential for effective lending decisions and financial stability in commercial banking. This study proposes an explainable machine learning framework for credit risk prediction in the context of commercial banks in Bangladesh. A synthetic borrower-level dataset consisting of 5000 records was developed to simulate plausible demographic, financial, loan-related, and repayment-behavior characteristics commonly associated with commercial banking practice in Bangladesh. Four machine learning models, namely Logistic Regression, Random Forest, Support Vector Machine, and XGBoost, were implemented and compared using accuracy, precision, recall, F1-score, and AUC-ROC. The results show that ensemble-based models outperform the baseline Logistic Regression model. XGBoost achieved the strongest overall classification performance, with an accuracy of 96%, precision of 0.95, recall of 0.94, and F1-score of 0.95, while Random Forest achieved the highest AUC-ROC value of 0.94. To improve transparency, SHAP and LIME were applied to provide global and local explanations of model predictions. The findings indicate that debt-to-income ratio, monthly income, loan amount, previous default history, and late payment behavior are the major drivers of predicted credit risk. Since the dataset is synthetic, the results should be interpreted as simulation-based evidence rather than direct empirical evidence from confidential commercial bank records. The proposed framework demonstrates the potential value of explainable machine learning for transparent and data-driven credit risk assessment in the Bangladeshi banking context.