A Hybrid Feature Selection Method for Data Sets of thousands of Variables

Feature selection has become the focus of research areas of applications with datasets of thousands of variables. In this study we present a hybrid feature selection (HFS) method that adopts both filter and wrapper models of feature subset selection. In the first stage of the feature selection, we use the filter model to rank the features by the mutual information (MI) between each feature and each class, and then choose k highest relevant features to the classes. In the second stage, we complete a wrapper model based feature selection algorithm, which uses Shepley value to evaluate the contribution of features to the classification task in a feature subset. Experimental results show obviously that the HFS method obtains better classification performance than solo Shepley value based or solo MI based feature selection method.