Segmentation of retail bank customers for the purposes of modeling the loan claim default
Subject. In the banking practice, approaches to separate modeling of loan claim default (for new and repeat customers, for customers having and not having a history in the Credit Bureau, etc.) are widespread. Such a segmentation of retail bank customers may increase the efficiency of applied scoring system. The practical problem of choosing the optimal heuristic method of segmentation for the scoring remains unresolved. Objectives. The purpose of this work is to determine the optimal heuristic method of segmentation from those that are known in the literature and the industry. Methods. The study employs statistical analysis and content analysis of information sources. Results. We compared over thirty heuristic methods for segmentation of retail bank customers. The comparison showed that according to the classifier of the efficiency metric (AUROC), our proposed segmentation by the disbursed loan size turned out to be optimal. The method consists in the ‘disbursed loan’ variable discretization under the TreeR method. Conclusions and Relevance. The findings may be helpful in loan scoring and in any statistical modeling, using the logit regression.