SVM margin-based feature elimination applied to high-dimensional microarray gene expression data

In this paper we investigate application of the recently developed margin-based feature elimination (MFE) method for feature selection in support vector machines to high-dimensional, small sample size data from the DNA microarray domain. We compared the performance of MFE to the well-known recursive feature elimination (RFE) method. Our results show that MFE outperforms RFE in terms of generalization accuracy and classifier margin, especially for low frequency of SVM retraining during the feature elimination process, which is practically necessitated for very high-dimensional feature spaces.