A Fast Clustering-based Feature Subset Selection Algorithm

The paper aims at proposing the fast clustering algorithm for eliminating irrelevant and redundant data. Feature selection is applied to reduce the number of features in many applications where data has hundreds or thousands of features. Existing feature selection methods mainly focus on finding relevant features. In this paper, we show that feature relevance alone is insufficient for efficient feature selection of high-dimensional data. We define feature redundancy and propose to perform explicit redundancy analysis in feature selection. A new hypothesis is introduced that dissociate relevance analysis and redundancy analysis. A clustering based method for relevance and redundancy analysis for feature selection is developed and searching based on the selected features will be performed. While the efficiency concerns the time required to find a subset of features, the effectiveness determines the quality of the subset of features. A fast clustering-based feature selection algorithm, FAST, has been selected to be used in the proposed paper. The clustering-based strategy has a higher probability of producing a subset of useful as well as independent features. To ensure the efficiency of FAST, efficient minimum-spanning tree clustering method has been adopted. When compared with FCBF, ReliefF, with respect to the classifier, namely, the tree-based C4.5, FAST not only produces smaller subsets of features but also improves the performances by reducing the time complexity. Keyterms: Clustering, Feature subset selection, Minimum Spanning Tree, T-Relevance, FCorrelation.