Hybrid feature selection through feature clustering for microarray gene expression data

Abstract

Goal of feature selection is to find a suitable feature subset that produces higher accuracy for classifier in the user end. Hybrid methods for feature selection comprised of combination of filter and wrapper approaches have recently been emerged as strong techniques for the problem in this domain. In this paper we have presented a novel approach for feature selection based on feature clustering using well known k-means philosophy for the high dimensional gene expression data. Also we have proposed three simple hybrid approaches for reducing data dimensionality while maintaining classification accuracy which combine our basic feature selection through feature clustering (FSFC) approach to other standard approaches of feature selection in different orientation. We have employed popular Box and Whisker plot and ROC curve analysis to evaluate experimental outcome. Our experimental results clearly show suitability of our methods in hybrid approaches of feature selection in micro-array gene expression domain.