Using Support Vector Machines for Multicategory Cancer Diagnosis Based on Gene Expression Data

In an effort to contribute to the development of accurate cancer diagnosis based on gene expression data, this study performs a comprehensive evaluation of multicategory Support Vector Machine (MC-SVM) algorithms applied to the majority of cancer-related gene expression microarray datasets currently freely available to the scientific community. Our results show that: (a) MC-SVMs are very effective in performing accurate cancer diagnosis in high-dimensional gene expression data. The MC-SVM techniques by Crammer and Singer, Weston and Watkins, and one-versus-rest are the best methods in this domain. (b) MC-SVMs outperform other popular machine learning algorithms to a remarkable degree. (c) A prototype software tool which develops MC-SVM classifiers in a fully-automated fashion is introduced. Results produced by the tool compare favorably with previously published studies.