Modelo de clasificación binaria para una población bancarizada empleando metodología de ensamble.

Abstract

In this project an empirical investigation of the performance of several classification models based on the methodology of homogeneous and heterogeneous ensemble for the prediction of the payment behavior of credit obligations acquired by individuals in a financial institution is carried out; the purpose is to reduce the cost of manual credit analysis, make earlier decisions and reduce the potential risk. The construction of more robust models with higher accuracy in the prediction of default is based on the combination of classifiers: Logistic Regression (LR), Random Forest (RF), Gradient Boosting Machine (GBM) and Naïve Bayes (NB). Additionally, the performance of the model results obtained with the ensemble methodology is superior to the performance of the results obtained with the individual classifiers.