A binary individual search strategy-based bi-objective evolutionary algorithm for high-dimensional feature selection

Abstract

Evolutionary computation is promising in tackling with the feature selection problem, but still has poor performance in obtaining good feature subset in high-dimensional problems. In order to efficiently obtain the optimal feature subset with higher classification accuracy and lower feature dimensions, a binary individual search strategy-based bi-objective evolutionary algorithm is proposed. The proposed algorithm has three advantages and contributions. Firstly, an improved fisher score is utilized to preprocess the feature space to remove the irrelevant and redundant features. It can decrease the feature dimensionality and compress the search space of feature subset effectively. Secondly, a binary individual search strategy is developed that contains a nearest neighbor binary individual crossover operator and an adaptive binary individual mutation operator, which can search the global optimal feature combination. Thirdly, enhanced population entropy and improved average convergence rate are adopted to monitor the correlation between the diversity of the population and the convergence of optimization objectives. Promising experimental results on twelve high-dimensional datasets reveal that the proposed algorithm can obtain competitive classification accuracy and effectively reduce the size of feature subset compared with ten state-of-the-art evolutionary algorithms.