Hybrid Model Based on Machine Learning for the Prediction of Consumer Credit Delinquency in the Banking Sector of Peru

Abstract

The delinquency rate among clients of banking institutions in Peru has increased exponentially in recent years, due to the lack of early detection of potentially delinquent clients, mainly due to the use of inadequate prediction techniques for the identification of delinquent clients. This causes profitability to be reduced, credit risk to increase and the country's economy to be unstable. Previously, different solutions were generated to prevent non-payment, however these studies were not applied in the Peruvian environment and did not cover the personal and financial variables necessary to improve the detection of delinquent clients. In this work, a delinquency prediction system is proposed using classification algorithms such as logistic regression and Random Forest, with the aim of improving and automating the early detection of delinquent clients and counteracting the increase in delinquency, so that banks can of Peru can reduce their financial losses due to non-payment by delinquent clients, and prevent the granting of consumer loans to clients who have a high probability of delinquency. After validating the performance of the algorithm using key indicators, it was obtained that the results are superior to the compared algorithms, thus showing a precision of 90 percent, a recall of 95 percent and an accuracy of 91 percent.