Credit Portfolio Assessment Using Machine Learning: A Case Study of a Mexican Non-Bank Financial Institution.

This study focuses on the development of a credit risk assessment model for non-bank financial institutions in Mexico, a segment that has historically been excluded from the traditional financial system and primarily serves marginalized populations with limited access to formal financial services. Given the inefficiency of conventional risk models in environments of high financial uncertainty, we implemented machine learning techniques using the R programming language, specifically Random Forest and XGBoost, to predict credit default. The analysis was based on a real sample consisting of 753 credit subjects, subjected to data transformation, normalization, and class balancing through the oversampling technique. Both models were evaluated using standard performance metrics such as accuracy, precision, recall, specificity, ROC AUC, and the Gini index. The evaluation showed that, although both models achieved an accuracy of 97.8%, XGBoost significantly outperformed Random Forest in identifying delinquent clients, standing out for its higher recall and AUC. These findings confirm the applicability of machine learning models as effective and robust tools for credit risk management in non-bank financial institutions. This study is limited by its small sample size, geographic restriction, and the potential impact of oversampling on its external validity. The absence of certain external variables is also a limitation. Nevertheless, these findings serve as an initial step toward developing more generalizable credit risk assessment approaches for non-bank financial institutions.