Colon Cancer Prediction from Gene Expression Profiles Using Kernel Based Support Vector Machine

Molecular omics (genomics, transcriptomics, proteomics, metabolomics, etc.) analysis is now playing the significant rule for cancer prediction, where gene expression profile is one of the most popular omicss. Kernel based Support Vector Machines (SVMs) are now widely using for cancer prediction based on the profiles of gene expression. However, the kernel based SVM's performance depends on both kernel and feature selection. Here, in this paper, we would like propose a way for both important kernel and features (important genes) selection for SVM to predict the colon cancer based on gene expression profile. Features (genes) selection has been done by t-statistic. Then a comparative study of cancer prediction accuracy of the proposed kernel based SVM with some popular predictors (Naive Bayes (NB), linear discriminate analysis (LDA) and quadratic linear discriminate analysis (QDA)) had been performed using the selected top 3 genes which are ranked with the help of the t-statistic. The proposed SVM-radial basis is found to be the better colon cancer predictor based on area under the curve (AUC) of the receiver operating characteristics and total accuracy. Thus we may reach to a conclusion that SVM-radial basis and t-statistic based feature selection altogether is an effective and feasible substitute to common techniques.