Improved Filter Method for Feature Selection

Abstract

To solve the problem of high data feature dimension in intrusion detection, a hybrid feature selection method has been proposed to reduce the feature dimension. This approach combines the filter and sequence floating forward search methods. Firstly, the original feature series is sorted by different filter methods, and the top ranked features are selected as the next original feature set. Based on this, the sequence floating forward search method is used to select the optimal feature subset with the Support Vector Machine (SVM) as the classifier. The method can avoid the selection and screening of the features based on the threshold and the evaluation value of single feature and categorical variable. Thereby the high evaluation characteristics obtained by the Filter method may be complementary to the low evaluation characteristics. The result shows that the proposed method can not only effectively reduce the number of features, but also can achieve a better classification performance.