A filter-dominating hybrid sequential forward floating search method for feature subset selection in high-dimensional space

Sequential forward floating search (SFFS) has been well recognized as one of the best feature selection methods. This paper proposes a filter-dominating hybrid SFFS method, aiming at high efficiency and insignificant accuracy sacrifice for high-dimensional feature subset selection. Experiments with this new hybrid approach have been conducted on five feature data sets, with different combinations of classifier and separability index as alternative criteria for evaluating the performance of potential feature subsets. The classifiers under consideration include linear discriminate analysis classifier, support vector machine, and K-nearest neighbors classifier, and the separability indexes include the Davies-Bouldin index and a mutual information based index. Experimental results have demonstrated the advantages and usefulness of the proposed method in high-dimensional feature subset selection.