A Feature Subset Evaluation Method Based on Multi-objective Optimization

To remove the irrelevant and redundant features from the high-dimensional data while ensuring classification accuracy, a supervised feature subset evaluation method based on multi-objective optimization has been proposed in this paper. Four aspects, sparsity of feature space, classification accuracy, information loss degree and feature subset stability, were took into account in the proposed method and the Multi-objective functions were constructed. Then the popular NSGA-II algorithm was used for optimization of the four objectives in the feature selection process. Finally the feature subset was selected based on the obtained feature weight vector according the four evaluation criteria. The proposed method was tested on 4 standard data sets using two kinds of classifier. The experiment results show that the proposed method can guarantee the higher classification accuracy even though only few numbers of features selected than the other methods. On the other hand, the information loss degrees of the proposed method are the lowest which demonstrates that the selected feature subsets of the proposed method can represent the original data sets best.