Feature Selection Via Fuzzy Clustering

This paper deals with feature selection for classification with wrapper framework. We develop a new algorithm for feature selection, based on a fuzzy clustering technique and an iterative process verifying classification accuracy. By monitoring discrepancy between two cluster systems, one derived with full features of the dataset, the other one with a subset of features, we are able to evaluate representation power of the subset of features with respect to the original feature set . Experimental results confirm efficiency of the proposed algorithm