Hybrid Hub erized Supp ort Vector Machines for Microarray Classification
Li Wang - Ross School of Business, University of Michigan, U.S.
Ji Zhu - Department of Statistics, University of Michigan, U.S.
Hui Zou - School of Statistics, University of Minnesota, U.S.
The large number of genes and the relatively small number of samples are typical characteristics for microarray data. These characteristics pose challenges for both sample classification and relevant gene selection. The support vector machine (SVM) is a widely used classification technique, and previous studies have demonstrated its superior classification performance in microarray analysis. However, a ma jor limitation is that the SVM can not perform automatic gene selection. To overcome this limitation, we propose the hybrid huberized support vector machine (HHSVM). The HHSVM uses the huberized hinge loss function and the elastic-net penalty. It has two ma jor benefits: 1. automatic gene selection; 2. the grouping effect, where highly correlated genes tend to be selected/removed together. We also develop an efficient algorithm that computes the entire regularized solution path for HHSVM. We have applied our method to real microarray data and achieved promising results.