Model for Predicting Success in Loan Recovery Using Machine Learning in a Peruvian Financial Institution

In an effort to improve loan recovery, Peruvian financial institutions have turned to predictive machine learning models to identify the loans with a higher repayment probability. This study proposes a model to predict the recovery of financial loans by using 3 machine learning algorithms: Random Forest, Logistic Regression, and Support Vectorial Machines. The CRIS-DM methodology was used considering the following steps: data compilation, data preprocessing, algorithm modeling, and metrics definition. The used dataset was collected from 438 historical loan records from a Peruvian financial company, from which 10 variables where identified. The Random Forest algorithm results stood out for its high precision (97.80%) and generalization capacity demonstrating better performance in predicting loan recovery. These findings highlight the viability of the “Random Forest” algorithm as a tool for improving loan credit risk management, and suggest it's implementation as part of an comprehensive credit evaluation system.