An Optimize Gene Selection Approach for Cancer Classification Using Hybrid Feature Selection Methods

Abstract

In the field of diseases diagnosis and treatment, microarray gene expression data plays a crucial role. But for analysis, expression data are available with a huge number of genes in comparison with few tissue samples. So, the most challenging task is to find out most influential genes from the high-dimensional, noisy and redundant microarray data. To overcome the above-said issues, in this paper, we proposed a two-stage gene subset selection mechanism by a combination of non-parametric Kruskal-Wallis test (KWs test) and Correlation-based Feature Selection (CFS) algorithms. The proposed technique selects most important and significant features (here genes) as well as eliminates insignificant and redundant features (here genes), that have been playing an important role to address this problem. Over three publicly available microarray datasets, proposed technique has been evaluated using two classifiers, namely supported vector machines (SVM) and k-nearest neighbors (k-NN). We also compared experimental outcomes obtained from our proposed model with recently published feature selection and classification models to determine whether or not proposed model is suitable for high-dimensional microarray data analysis. The proposed technique achieves the prediction accuracy rate of 98.61% for leukemia, 90.90% for colon cancer, and 99.60% for ovarian cancer using a support vector machine (SVM). Compared to other existing models, our proposed model shows relatively higher accuracy. Therefore, the proposed model can be used as a reliable framework for gene selection in cancer classification.