Score crediticio de Pymes en Banco Nacional de Costa Rica
Fecha
2025-03-26
Autores
Quirós Muñoz, Tatiana
Título de la revista
ISSN de la revista
Título del volumen
Editor
Resumen
Esta investigación tiene como objetivo desarrollar un score crediticio para PYMES en el Banco Nacional de Costa Rica, que permita clasificar a los clientes que soliciten un cédito por primera vez como buenos o malos pagadores. Para ello, se implementaron y compararon diferentes métodos de clasificación de Machine Learning. Los resultados indican que el modelo más competitivo en términos de precisión, interpretabilidad y explicabilidad es el método GAMI-Net. Además, este método permitió garantizar la solidez y efectividad del diseño del modelo a través de un diagnóstico exhaustivo. Los hallazgos de esta investigación sugieren que los desarrolladores de modelos de Machine Learning deberían apoyarse en herramientas que faciliten la comprensión del proceso de modelado y la interpretación de los resultados, así como en herramientas de diagnóstico post-entrenamiento para identificar falencias y posibles mejoras en los modelos.
This research aims to develop a credit scoring model for SMEs at the Banco Nacional de Costa Rica, allowing the classification of first-time loan applicants as good or bad payers. To achieve this, various Machine Learning classification methods were implemented and compared. The results indicate that the most competitive model in terms of accuracy, interpretability, and explainability is the GAMI-Net method. Furthermore, this method ensured the robustness and effectiveness of the model design through a thorough diagnostic analysis. The findings of this research suggest that Machine Learning model developers should rely on tools that facilitate the understanding of the modeling process and the interpretation of results, as well as post-training diagnostic tools to identify weaknesses and potential improvements in the models.
This research aims to develop a credit scoring model for SMEs at the Banco Nacional de Costa Rica, allowing the classification of first-time loan applicants as good or bad payers. To achieve this, various Machine Learning classification methods were implemented and compared. The results indicate that the most competitive model in terms of accuracy, interpretability, and explainability is the GAMI-Net method. Furthermore, this method ensured the robustness and effectiveness of the model design through a thorough diagnostic analysis. The findings of this research suggest that Machine Learning model developers should rely on tools that facilitate the understanding of the modeling process and the interpretation of results, as well as post-training diagnostic tools to identify weaknesses and potential improvements in the models.
Descripción
Palabras clave
Machine learning, Score de crédito, caja negra, red neuronal, GAMI-net