Hybridized KNN and SVM for gene expression data classification

Support vector machine (SVM) is one of the most powerful supervised learning algorithms in gene expression analysis. The samples intermixed in another class or in the overlapped boundary region may cause the decision boundary too complex and may be harmful to improve the precise of SVM. In the present paper, hybridized k-nearest neighbor (KNN) classifiers and SVM (HKNNSVM) is proposed to deal with the problem of samples in the overlapped boundary region and to improve the performance of SVM. The first KNN is used to prune training samples and the second KNN is combined with SVM to classify the cancer samples. The proposed algorithm was used in binary and multiclass classification of gene expression data. The results were compared to those obtained by single SVM and KNN. It has been demonstrated that the proposed method is a useful tool for classification and the misclassification rate for the prediction set is reduced with samples pruning used. Compared with SVM and KNN, the misclassification rates of HKNNSVM for the datasets containing mislabeled samples were notably lower than that by SVM and KNN, which indicated that the classification performance of HKNNSVM was stable. [Life Science Journal. 2009; 6(1): 61 – 66] (ISSN: 1097 – 8135).