A Feature Selection Algorithm Based on Equal Interval Division and Conditional Mutual Information

Xiangyuan Gu, Jichang Guo, Tao Ming, Lijun Xiao, Chongyi Li

Abstract: The performance of many feature selection algorithms is affected because of ignoring three-dimensional mutual information among features. Three-dimensional mutual information includes conditional mutual information, joint mutual information and three-way interaction information. Aiming at the limitation, this paper investigates feature selection based on three-dimensional mutual information. First, we propose an objective function based on conditional mutual information. Further, we propose a criterion to validate whether the objective function can guarantee the effectiveness of selected features. In the case that the objective function cannot guarantee the effectiveness of selected features, we combine a method of equal interval division and ranking with the objective function to select features. Finally, we propose a feature selection algorithm named EID-CMI. To validate the performance of EID-CMI, we compare it with several feature selection algorithms. Experimental results demonstrate that EID-CMI can achieve better feature selection performance.