A fusion approach based on wrapper and filter feature selection methods using majority vote and feature weighting

An essential issue in machine learning is identifying a representative features set that serve to construct robust classification models for a particular task. This paper addresses the problem of feature selection for credit scoring. Many feature selection methods were proposed in the past few years to construct accurate credit scoring model. Instead of using a single feature selection method, we propose a new fusion approach that utilizes a number of diverse feature selection methods to do the job. Many approaches have been proposed to build such accurate credit model, of which filters and wrappers were the most popular. A fusion of this two approaches, is thoroughly investigated in this work and several comparisons are carried out to compare the proposed fusion method with the individual approaches. Evaluations on two credit datasets show that feature subsets selected by the fusion method improves classification accuracy.