Abstract:

The aim of this article is to assess the potential use of machine learning methods for the early identification of debt collection cases with low recovery potential in the management of mass receivables portfolios. The early identification of such cases is important from both managerial and economic perspectives, as it can support the prioritisation of actions, the reduction of ineffective costs, and the more efficient use of the operational resources of a debt collection entity. The study was conducted on a real dataset of 201,595 cases obtained from a debt collection entity. The target variable was defined as information on whether the total amount of payments was lower than the purchase price of the receivable. Three classification models were compared: logistic regression, Random Forest, and XGBoost. The best results were achieved by the XGBoost model. The results indicate that tree-based models identify low-potential cases more effectively than the linear model. In addition, SHAP analysis was applied, which increased the interpretability of the model and made it possible to link predictive results with expert knowledge. The article shows that machine learning can support portfolio segmentation and case prioritisation, but it should serve as a decision-support tool rather than a tool for automating debt collection decisions.