Statistical Analysis of Consumer Credit Risks Using Machine Learning Methods

This paper is devoted to determining the optimal tools for modeling the risk of early repayment of unsecured consumer loans. The first chapter introduce a brief history of the development and regulation of the consumer lending market, as well as a survival analysis methodology. The second chapter presents the modelling the risk functions of early repayment of unsecured consumer loans using traditional statistical lifetime models and approaches based on machine learning methods. The results showed that for the studied loan portfolio of an international credit institution, the use of machine learning models, such as Random Survival Forest and Gradient Survival Boosting, can improve the efficiency of early repayment risk assessment in comparison with traditional statistical approaches. The proposed models allow to obtain interpreted results of higher accuracy, to take into account the nonlinearity of covariates, to avoid miss-specification and inconsistent estimates.