Microarray Gene Expression Cancer Diagnosis Using Multiclass Support Vector Machines

Cancer diagnosis is one of the most important emerging clinical applications of gene expression microarray technology. In this paper, the global k-means algorithm was adopted to cluster genes in training samples sets. The gene with largest Dudoit ratio in each cluster was picked as a characteristic gene. A multiclass support vector machine (SVM) was developed as a classifier for the prediction of cancer types. With a test on LC197 data set and TC168 data set, the method in this paper not only achieved better performances than One-vs-One SVM and One-vs-Rest SVM, but also outperformed k-nearest neighbors and artificial neural network. These results render this method a viable alternative to other classification methods for cancer diagnosis.