Tumor diagnosis with support vector machines

This paper presents the application of SVMs to gene expression data based tumor diagnosis. Since there are large amount of genes and small number of samples in gene data and too many genes can harm the performance of the discrimination and increase the cost as well, a novel gene selection method is also proposed. Compared with the well-known Fisher algorithm on two open data sets, SVMs show higher performance. The significances of kernel function, soft margin parameter of SVM and gene selection are also discussed in this paper.