Abstract

This article discusses how to apply the sampling method called “Cube method” in credit scoring problems in order to improve the precision of the obtained predictive models. This method allows to guarantee an optimal balance of the samples when working with databases whose classes of the dependent variable are highly unbalanced. Using two samples of real bank data, a comparative study of the best models obtained with various data mining methods applied to the original databases is carried out against the balanced ones. Finally, it is concluded that the predictive capacity of the classification algorithms is more precise and that the models used reduce the economic cost of classification when the samples are balanced.