An Assertive Machine Learning Model for Rural Micro Credit Assessment in Peru

Abstract

The rural population in Peru has limited access to the financial system due to the high cost of credit and the high risk (default rates) caused by the informal sector. Thus, it is necessary to improve the assertive microcredits loans in favor of the rural economy. The Peruvian model is based mainly on evaluation achieved by experts, whose main tasks are to evaluate and verify clients requesting these microcredits. Currently, these tasks are performed manually and subjective to the expert’s judgment. This work proposes to find the highest level of assertiveness for the credit granting process and the consequent reduction of credit risk using several Machine Learning models. These models considered significant variables of the microcredit evaluation process in rural areas, testing techniques such as SMOTE and K-fold and, evaluating the models using some metrics, such as Accuracy, Precision, Recall, F1 Score, AUC ROC. The LightGBM model, based on decision trees, achieved an excellent level of assertiveness, with a 96.20% loan success rate. The results to reduce the delinquency rate prove that it is optimal to use technological tools such as machine learning models to support decision-making by experts of credit risk assessment in rural areas.