Intelligent decision support for debt collection using predictive learning and multi-criteria optimization

An efficient debt collection process should aim not only to recover maximum amounts but also to minimize operational costs. This paper proposes a novel approach to optimizing debt collection, by integrating Machine Learning (ML), Fuzzy Logic, and Multi-Criteria Decision Analysis (MCDA). Our framework consists of three layers: Rule Extraction, Prediction, and Optimization. The Rule Extraction layer employs unsupervised learning to identify customer segments (e.g., self-cured, lazy payers, and delinquents), thus enabling tailored strategies for each group. The Prediction layer uses supervised learning to forecast customer behavior, refining outreach methods. The Optimization layer uses these insights to balance costs with recovery rates. A key feature of our approach is the use of Fuzzy Logic to interpret ML outputs and create comprehensive risk profiles. Additionally, MCDA refines decision-making by considering both financial goals and practical constraints. We demonstrate the framework’s effectiveness using anonymized mortgage loan data provided by a debt collection agency managing portfolios on behalf of a commercial bank. Our solution approach addresses challenges like predicting payment outcomes and improving collection strategies. Computational experiments show high accuracy in predictions, and the Optimization layer illustrates how to allocate field collectors optimally, balancing recovery rates with operational costs. • Addresses debt collection as a holistic decision problem rarely explored in literature. • Proposes a unified framework integrating ML, fuzzy logic, and MCDA. • Accurately predicts customer behavior and payment outcomes. • Optimizes resource allocation balancing recovery rates and operational costs. • Validated through real-world computational experiments.