Chaotic binary bat algorithm with mutual information for feature subset selection

This paper presents an improved method for selecting the best features for data, based on the combination of the mutual information (MI) method and the chaotic binary bat algorithm (CBBA). The proposed method, named MI-CBBA, is based on three stages: (1) MI is used to rank the most relevant features in order of importance from the highest to the lowest importance, (2) a chaotic sine map is used to generate the initial population parameters for the binary bat algorithm, and (3) the binary bat algorithm is applied as an additional stage to reduce the dimensionality of the data and obtain the best features. The results obtained through application to biological data show that the proposed MI-CBBA algorithm has higher classification accuracy with a smaller number of selected features compared to the standard bat algorithm.