<![CDATA[Microfinance institutions face substantial challenges in managing financing risk, particularly in assessing the creditworthiness of mustahik when available data are limited. BAZNAS Microfinance Desa (BMD) requires a predictive risk system that is both accurate and transparent to ensure program sustainability while adhering to sharia principles. This study develops an Explainable Boosting Machine (EBM) model using historical data from 736 mustahik across three BMD locations (2019-2024). The methodology integrates comprehensive feature engineering, including the DTI Ratio, Savings Ratio, Financial Stress Indicator, and Dependency Ratio. Model performance was evaluated using ROC-AUC, precision-recall metrics, and confusion matrix analysis, while interpretability was examined through SHAP values and partial dependence plots. The EBM model achieved strong predictive performance, recording an ROC-AUC of 0.853, an accuracy of 80%, a precision of 82%, and a recall of 77%. Global interpretability analysis identified Remaining Balance (18.2%), Business Type (12.5%), and Household Income (11.3%) as the most influential predictors. Feature-engineered variables contributed 42% to the model’s predictive strength, confirming the added value of domain-knowledge-driven feature engineering. Critical risk thresholds were identified at Remaining Balance below IDR 200,000 and DTI Ratio above 0.8. The EBM framework effectively balances predictive accuracy with full interpretability, making it suitable for deployment in microfinance decision-support systems. The model provides actionable insights for risk-based pricing and early warning mechanisms while maintaining the transparency essential in microfinance financing.]]>
Copyrights © 2025
