Gene Selection for Cancer Classification Using Relevance Vector Machine

How to select best gene subset out of thousands of genes in microarray data is important for accurate cancer classification. SVM-RFE is one of the important approaches for gene selection, which combines support vector machine with recursive feature elimination, and the method ranks the genes with recursive procedure. A novel machine learning method called relevance vector machine is proposed by Tipping in 2000, as an alternative and direct competitor to the SVM. In this paper, we propose RVM-RFE method for gene selection by combining RVM and RFE. Compared to the SVM-RFE, the experiments on the real datasets suggest that RVM-RFE can lead to comparable Loocv accuracy and shorter running time; further research suggests that our method is also much better than linear RVM and other popular methods.