Feature Selection via Normalized Dynamic Change of Selected Feature with Class

Abstract

Feature selection has been widely used in various application areas such as machine learning, bioinformatics, and natural language processing. Common drawbacks of most of the current feature selection methods are the lack of information about the dynamic change of selected features with the class, and the selection of redundant and irrelevant features. In this paper, we develop a novel feature selection method called Normalized Dynamic Change of Selected Feature with Class (NDCSF), which consider the normalized dynamic information changes between the selected features and the classes by using conditional mutual information and entropy. Moreover, a normalized feature redundancy by using mutual information and entropy is introduced into NDCSF. The experimental results on several benchmark datasets verify that the NDCSF can significantly improve the other several feature selection methods.