Feature selection using feature ranking, correlation analysis and chaotic binary particle swarm optimization

In this paper, we propose a multi-stage feature selection algorithm, which focuses on the reduction of redundant features and the improvement of classification performance using feature ranking (FR), correlation analysis (CA) and chaotic binary particle swarm optimization (CBPSO). In the first stage, with the purpose of selecting the most effective features for classification, FR is introduced to select the top-ranked features according to the classification accuracies. In the second stage, CA is used to measure the correlation among the selected top-ranked features for reducing redundant features. In the third stage, in order to further eliminate redundant features and improve the classification performances, CBPSO is adopted to search the optimal feature subset. Ultimately, feature selection can be completed by using only some top-ranked features with less redundancy for classification. Support vector machine (SVM) with n-fold cross-validation is adopted to assess the classification performances on six datasets in the experiments. Experimental results show that the proposed algorithm can achieve better performance in terms of classification accuracy and the number of features than benchmark algorithms.