<![CDATA[Credit risk remains a critical concern for rural banks because lending quality directly affects financial stability and institutional sustainability. Although banking databases contain valuable debtor information, credit assessments often continue to rely on administrative verification and subjective judgment. This study aimed to identify credit risk characteristics, compare the performance of Random Forest, Gradient Boosted Tree, and Random Tree algorithms, and formulate data-driven risk management strategies for PT BPR Jabar Perseroda. A quantitative descriptive research design was employed using the Cross-Industry Standard Process for Data Mining (CRISP-DM) framework, integrating descriptive, predictive, and prescriptive analytics. The sample consisted of 1,015 debtor records selected through stratified sampling from a population of 10,150 records. Model performance was evaluated using accuracy, precision, and recall metrics, while correlation analysis was conducted to identify variables associated with credit risk status. Payment delinquency history showed the strongest relationship with risk status (r = 0.653), whereas collateral demonstrated the weakest relationship (r = -0.047). The Random Forest algorithm achieved the best predictive performance, with 99.00% accuracy, 100.00% precision, and 97.83% recall. The findings indicated that repayment behavior provided more meaningful risk information than static administrative attributes. Therefore, integrating Random Forest-based predictions with the 5C credit assessment principles could strengthen objective credit evaluation, decision-support systems, early warning mechanisms, and proactive credit risk management.]]>
Copyrights © 2026
