Abstract:

The article presents the application of unsupervised clustering methods in the management of mass debt portfolios, with particular emphasis on the K-Modes algorithm. The aim of the study was to identify homogeneous groups of receivables based on selected characteristics of both receivables and debtors, and to assign them differentiated collection strategies in order to maximize the efficiency of debt recovery. The analyses showed significant differences in debt profiles, repayment levels, and the effectiveness of collection activities undertaken between the identified clusters. Strategies assigned by industry experts were adjusted to the specifics of each group, allowing for an increase in recoveries while simultaneously minimizing the operational costs associated with the mass debt collection process. The research results also indicate the possibility of using the developed clustering models for the automatic assignment of new debt portfolios to appropriate segments, which significantly increases portfolio management efficiency in dynamically changing market conditions. The integration of data mining techniques with expert knowledge has been confirmed as an effective way to increase the effectiveness of collection processes undertaken in a specialized debt collection entity and may constitute a basis for further automation and personalization of the undertaken activities.