<![CDATA[This study examines the influence of Days Past Due, Loan Amount, and Savings Proportion on credit default risk in CU Sejahtera Makmur Bersama using a comparative machine learning approach. Random Forest and XGBoost models were implemented and evaluated using Mean Squared Error (MSE), Mean Absolute Error (MAE), and the Coefficient of Determination (R²). The results indicate that XGBoost_Optimized achieved the best performance with the lowest MSE (23,124,275,951) and the highest R² (0.020977), while Random Forest showed slightly better MAE performance (69,923,891). However, the R² value remains very low, indicating that the models explain only around 2% of the variance in credit default outcomes. This limited explanatory power is likely attributable to the absence of non-financial and behavioral variables, such as borrower character, repayment discipline, and employment stability, which are critical determinants of default behavior in cooperative lending. Additionally, potential data imbalance or outlier effects may have further reduced predictive accuracy. These findings suggest that financial indicators alone are insufficient to capture the complexity of non-performing loans. Future research should integrate borrower behavioral attributes, macroeconomic variables, and alternative modeling strategies such as binary classification or risk segmentation techniques to improve predictive performance and practical applicability.]]>
Copyrights © 2026
