<![CDATA[Debt collection generates extensive interaction data collection actions, contact outcomes, and communication history that is traditionally recorded only qualitatively for documentation, limiting its analytical value. This study aims to develop a scoring-based framework that converts qualitative collection data into quantitative indicators for data-driven decision-making. The method begins with a rule-based scoring mechanism to encode collection actions and interaction history into a total score reflecting customer payment potential, then extends this baseline using machine learning to learn feature importance and improve prediction accuracy. Model performance is evaluated using AUC ROC and Precision@K for payment likelihood, MAE and RMSE for payment time prediction, and recovery rate for business impact. Results show that the hybrid model significantly improves prioritization efficiency and recovery outcomes compared to conventional qualitative methods, demonstrating how operational data can be systematically quantified and integrated with predictive analytics for smarter collection strategies.]]>
Copyrights © 2026
