Semi-supervised SVM-based Feature Selection for Cancer Classification using Microarray Gene Expression Data

Gene expression data always suffer from the high dimensionality issue, therefore feature selection becomes a fundamental tool in the analysis of cancer classification. Basically, the data can be collected easily without providing the label information, which is quite useful in improving the accuracy of the classification. Label information usually difficult to obtain as the labelling processes are tedious, costly and error prone. Previous studies of gene selection are mostly dedicated to supervised and unsupervised approaches. Support vector machine (SVM) is a common supervised technique to address gene selection and cancer classification problems. Hence, this paper aims to propose a semi-supervised SVM-based feature selection (SVM-FS), which simultaneously exploit the knowledge from unlabelled and labelled data. Experimental results on the gene expression data of lung cancer show that SVM-FS achieves the higher accuracy yet requires shorter processing time compares with the well-known supervised method, SVM-based recursive feature elimination (SVM-RFE) and the improved method, SVM-RFE.