Identification and Interpretation of Significant Factors Influencing Client Defaults in Microfinance Institutions Using Machine Learning Methods

One of the problems of business analytics is the lack of understanding by decision makers of the process of obtaining results presented by machine learning methods. The paper discusses the background and results of a study aimed at discovering factors that significantly affect the risk of non-repayment of loans by clients of microfinance institutions. The issues of application of gradient bousting and SHAP methods to identify factors characterizing microcredit borrowers are considered. The results of the research show that the interpretation of complex models using the SHAP method makes it possible to apply advanced machine learning methods to obtain analytical recommendations in the field of credit risk management in microfinance institutions. #CSOC1120.