Identification of the bank’s default clients by machine learning methods on the basis of binning

Abstract

Introduction. In the current global crisis, the problem of the quality of banks’ loan portfolios is a topical issue. Among the methods of effective credit risk management is the assessment of the borrower’s creditworthiness. Improving the quality of analysis of the strengths and weaknesses of the counterparty will reduce the occurrence of unforeseen risks in the process of conducting credit operations. Given the importance of the role of creditworthiness assessment for decision-making, there is a need to improve and choose a methodology that will ensure the most accurate classification of the bank’s clients.

Purpose. The aim of the work is to choose the best method for predicting the probability of default of commercial bank customers based on the analysis of approaches and testing of the built models.

Method (methodology). The paper considers methodological approaches to modeling the insolvency of bank customers and determining the probability of repayment of loans based on binning indicators. Also, the credit risk assessment models based on the use of logit and probit regressions, the algorithm of extreme gradient boosting and artificial neural networks are constructed. The comparative analysis of the efficiency of the application of the used approaches is carried out.

Results. The obtained results demonstrated the high accuracy of the models and their ability to identify non-creditworthy customers. The findings of the study and evaluation of mathematical approaches can be implemented in the work of banking structures and other credit institutions to spread the amount of problem fees in their loan portfolios.