Diagnostic prediction of multi-class cancer using SVM and nearest neighbor classifier

Abstract

Precise diagnosis of four heterogeneous childhood cancers, namely, neuroblastoma, non-Hodgkin lymphoma, rhabdomyosarcoma and Ewing sarcoma is crucial because they present a similar histology of small round blue cell tumors (SRBCTs) and frequently leads to misdiagnosis. However, due to small number of samples compared to very large number of genes in microarray gene expression data, it is hard to identify a small subset of relevant genes that can classify these four subgroups of childhood cancers with high accuracy. Therefore, in this paper, we have utilized t-test to rank all the genes according to their importance. Support vector machine (SVM) with different kernels and a simple 1-nearest neighbor (1-NN) classifier have been used to perform the classification task. Results demonstrate that the method could find very few numbers of genes for the diagnostic prediction of cancer subgroups.