Comparison of feature selection approaches based on the SVM classification

The credit scoring has been regarded as a critical topic. Creating an effective classificatory model will objectively help managers instead of intuitive experience. This study proposed four strategies combining with the SVM (support vector machine) classifier for features selection that retains sufficient information for classification purpose. Different features preprocessing steps were constructed with four strategies of conventional Linear discriminate analysis (LDA), Decision tree, Rough set and F-score models to optimize feature space by removing both irrelevant and redundant features. The accuracy of four models are compared and nonparametric Wilcoxon signed rank test was held to show the significant difference between these models. Our results suggest that hybrid credit scoring models can mostly classify the applicants as either good or bad clients that are robust and effective in finding optimal subsets and are a promising method to the fields of data mining.