Abstract
This paper gives a contribution in variable identification within credit scoring models using Random Forest. Specifically, we provide some insights about the behavior of the variable importance index based on random forests, focusing on the differences between “for-profit” and “not-for-profit” enterprises. We investigate two classical issues of variable selection: the first one is variable extraction for bankruptcy interpretation, whereas the second one is more restrictive and tries to design a good prediction model. Finally we provide an application to a real data set provided by Banca Popolare Etica.