Comparación de un método clásico (logit) y dos con técnicas de boosting (adaboost, gradient boosting) para la clasificación crediticia en una entidad financiera ecuatoriana.

Abstract

In this research project, referees of three classification methods: logistic regression, adaboost and gradient boosting. In the credit scoring modeling process is based on a data provided by a financial institution in Ecuador dedicated a credit portfolio of unknown size. The three developed models were built with R and Answer Tree v3.0 programs. Once established, was compared its predictive ability and classify clients classified as good and bad. Model performances were calculated for the Area Under Receiver Operator Characteristic Curve (AUC-ROC), Kolmogorov-Smirnov (KS) and Gini statistics. the error and success rate of each model were evaluated.