A grouping feature selection method based on feature interaction

Abstract

Feature interaction is crucial in the process of feature selection. In this paper, a grouping feature selection method based on feature interaction (GFS-NPIS) is proposed. Firstly, a new evaluation function measuring feature interaction is proposed. Secondly, a grouping strategy based on approximate Markov blanket is used to remove strong redundant features. Lastly, a new feature selection method called as GFS-NPIS is given. In order to verify the effectiveness of our method, we compare GFS-NPIS with other eight representative ones on three classifiers (SVM, KNN and CART). The experimental results on fifteen public data sets show that GFS-NPIS outperforms others in terms of classification accuracy and Macro-F1.