Feature subset selection using Information Energy and correlation coefficients of hesitant fuzzy sets

Abstract

In this paper, a novel feature selection algorithm based on hesitant fuzzy sets (HFS) is proposed. For each feature two HFSs are defined.For generating the first HFS for each defining feature, the opinions of three different ranking algorithms are considered. For second HFS for each feature the opinions of three different proximity measures are considered. The Information Energy (IE) of the first HFS for each feature is considered as the relevancy measure of the feature to the class labels. Then the hesitant correlation coefficient matrix for features is calculated based on the second HFSs. After that the average of hesitant correlation coefficients is considered as the relevancy measure of selected features. By combining hesitant based relevancy and redundancy measures, a new feature selection merit is proposed. The proposed merit potentially is able to consider both the maximum relevancy and the minimum redundancy of selected features. The efficiency of this approach is proved through 9 UCI repository datasets. The approach demonstrates a significant performance in both number of selected features and classification accuracy by four different classifiers.