An Improved Fuzzy Feature Clustering and Selection based on Chi-Squared-Test

Feature selection aims to reduce the dimensionality of patterns for classification by selecting the most informative instead of irrelevant and/or redundant features. In this paper, fuzzy feature clustering is proposed for grouping features based on their interdependence and selecting the best one from each cluster. Different novel fuzzification techniques for selection step are also introduced. Applying chi-square test, this approach considers the dependence of each feature on class labels during selection. Hence, it leads to remove redundant clusters of features which are unrelated to the class labels. The proposed method has two advantages. Firstly, it has more stability and faster convergence due to fuzzy clustering; secondly, it improves the accuracy of the classifier using the selected features. Experimental results demonstrate the good performances of this method on UCI benchmark data sets.