Predicción de mora en préstamos de consumo en la cooperativa CAYCSOL, utilizando algoritmos de machine learning, basada en datos históricos del periodo 2021–2024
No Thumbnail Available
Date
2026-01-01
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad Tecnológica Centroamericana UNITEC
Abstract
El presente Trabajo Final de Graduación tuvo como propósito desarrollar y validar un modelo predictivo de morosidad aplicado a los préstamos de consumo de la Cooperativa de Ahorro y Crédito Sonaguera Limitada (CAYCSOL), con el fin de fortalecer la gestión institucional del riesgo crediticio. La investigación se orientó a mejorar la identificación oportuna de clientes con probabilidad de incurrir en mora igual o superior a 30 días, mediante el análisis de información histórica sociodemográfica, financiera y crediticia correspondiente al período 2021–2024. Metodológicamente, el estudio se desarrolló bajo un enfoque cuantitativo, con un alcance descriptivo-predictivo y un diseño no experimental, incorporando análisis exploratorio de datos, selección de variables, técnicas de balanceo de clases y la evaluación comparativa de distintos modelos de Machine learning. Los resultados evidenciaron que el modelo Random Forest presentó el mejor desempeño predictivo, destacándose por su mayor capacidad para identificar clientes de alto riesgo y por la reducción de los falsos negativos. Como conclusión principal, se determinó que la aplicación del modelo predictivo aporta un valor significativo a la gestión del riesgo crediticio de CAYCSOL, por lo que se recomienda su implementación operativa y el establecimiento de mecanismos de monitoreo continuo que aseguren su efectividad y sostenibilidad en el tiempo.
The purpose of this Final Graduation Project was to develop and validate a predictive model of delinquency applied to consumer loans of the Sonaguera Savings and Credit Cooperative Limited (CAYCSOL), with the aim of strengthening the institution’s credit risk management. The study focused on improving the timely identification of clients with a probability of incurring delinquency equal to or greater than 30 days, through the analysis of historical sociodemographic, financial, and credit data corresponding to the 2021–2024 period. Methodologically, the research followed a quantitative approach, with a descriptive–predictive scope and a non-experimental design, incorporating exploratory data analysis, variable selection, class balancing techniques, and a comparative evaluation of different Machine learning models. The results showed that the Random Forest model achieved the best predictive performance, standing out for its greater ability to identify high-risk clients and for reducing false negatives. As the main conclusion, it was determined that the implementation of the predictive model provides significant value to CAYCSOL’s credit risk management, and its operational adoption is recommended, along with the establishment of continuous monitoring mechanisms to ensure its effectiveness and sustainability over time.
The purpose of this Final Graduation Project was to develop and validate a predictive model of delinquency applied to consumer loans of the Sonaguera Savings and Credit Cooperative Limited (CAYCSOL), with the aim of strengthening the institution’s credit risk management. The study focused on improving the timely identification of clients with a probability of incurring delinquency equal to or greater than 30 days, through the analysis of historical sociodemographic, financial, and credit data corresponding to the 2021–2024 period. Methodologically, the research followed a quantitative approach, with a descriptive–predictive scope and a non-experimental design, incorporating exploratory data analysis, variable selection, class balancing techniques, and a comparative evaluation of different Machine learning models. The results showed that the Random Forest model achieved the best predictive performance, standing out for its greater ability to identify high-risk clients and for reducing false negatives. As the main conclusion, it was determined that the implementation of the predictive model provides significant value to CAYCSOL’s credit risk management, and its operational adoption is recommended, along with the establishment of continuous monitoring mechanisms to ensure its effectiveness and sustainability over time.
Keywords
Aprendizaje Automático, Morosidad, Préstamos de Consumo, Riesgo Crediticio, Random Forest