Simultaneous gene selection and weighting in nearest neighbor classifier for gene expression data

Abstract

Gene expression data, such as microarray data, plays an important role as a biomarker in order to help in the effective cancer diagnosis, tumor classification or drug design at molecular level. However, due to highdimensionality of microarray datasets, they tend to have irrelevant or redundant features, and may lead to poor classification performance. For this reason, feature selection methods are commonly used to reduce the amount of data and to select relevant genes to improve the accuracy of machine learning methods. In this paper, a simultaneous feature ranking and weighting gene selection method with a nearest neighbor-based classifier is presented. In order to demonstrate the effectiveness of this proposal, a range of experiments over four well-known microarray datasets were carried out. Results showed that our method outperforms previous methods in terms of classification accuracy. Furthermore, evidence for significance of our results by means of non-parametric Friedman test is provided.