Classification of High-Risk Provinces for Fintech Lending Based on TWP90 from LPBBTI OJK Using Interpretable Model for Artificial Intelligence Ethical Transparency

Abstract

The growth of the fintech lending industry (LPBBTI) in Indonesia has expanded access to financing but has also increased credit risk, as reflected in the 90-day Default Rate (TWP90). This condition demands proactive regional risk monitoring, while existing approaches are still dominated by historical descriptive analysis. This study aims to develop a high-risk provincial classification as a transparent and accountable early warning system. The research methodology uses multi-sheet data integration from the Financial Services Authority (OJK) LPBBTI Statistics to create a provincial-monthly panel dataset covering supply, demand, and transaction activity. The Logistic Regression model was used as the baseline model due to its interpretability and support for decision auditability. Model evaluation using a time-based split approach during the May–July 2025 test period demonstrated good performance, with an ROC-AUC of 0.8598, an accuracy of 0.8462, and a precision of 0.8000. The results of feature analysis indicate that scale and activity indicators, particularly the outstanding amount, the number of active borrowers, and the value of disbursed funds, contribute significantly to risk probability. Although detection sensitivity (recall: 0.5333) still needs improvement, this study provides a measurable, relevant regional risk-ranking framework for regulatory decision-making.