A hybrid approach to feature ranking for microarray data classification

Abstract

We present a novel approach to multivariate feature ranking in context of microarray data classification that employs a simple genetic algorithm in conjunction with Random forest feature importance measures. We demonstrate performance of the algorithm by comparing it against three popular feature ranking and selection methods on a colon cancer recurrence prediction problem. In addition, we investigate biological relevance of the selected features, finding functional associations of corresponding genes with cancer. © Springer-Verlag Berlin Heidelberg 2013.