An ensemble wrapper feature selectionfor credit scoring

Abstract

In this paper, we address the problem of credit scoring (CS) as a feature selection problem. More specifically, we use wrapper feature selection methods to identify features that contain the most relevant information to distinguish good loan applicants from bad loan applicants. Wrapper feature selection approaches are widely used to select a small subset of relevant features from a dataset. However, wrappers suffer from the fact that they only use a single classifier in the evaluation process and each classifier is of a different nature and will have its own biases. Hence, this paper investigates the effects of using different classifiers for wrapper feature selection. A new ensemble method for feature selection is then proposed and evaluated on four credit datasets, and results illustrate that combining classifiers improves the performance of scoring models.