<![CDATA[The motorcycle-leasing sector in Indonesia is critical for consumer financing, yet firms face persistent difficulty in predicting early installment defaults because most credit-risk models are built for single organizations, forcing companies to repeatedly rebuild models when policies or operations change, which increases costs and delays risk detection. This study examines whether early default prediction models developed in one motorcycle-leasing company can be transferred to others by applying a hierarchical framework that integrates feature engineering, behavioral clustering, and supervised classification. The model was trained on 113,222 fiduciary contracts (2011–2025) from a Batam-based firm and tested on two external firms in Batam and Jakarta using Logistic Regression, Random Forest, and LightGBM. Results show substantial performance decline for the second Batam firm but relatively stable performance in Jakarta, indicating that organizational policy differences matter more than regional factors. Fine-tuning with sufficient local data improves performance, while limited data creates instability. The study provides a practical foundation for scalable and transferable credit-risk modeling in emerging markets. Keywords: credit risk prediction; transfer learning; domain adaptation; motorcycle leasing; clustering.]]>
Copyrights © 2026
