Molecular Diagnosis of Tumor Based on Independent Component Analysis and Support Vector Machines

Gene expression data that is being used to gather information from tissue samples is expected to significantly improve the development of efficient tumor diagnosis. For more accurate classification of tumor, extracting discriminant components from thousands of genes is an important problem which becomes challenging task due to the large number of genes and small sample size. We propose a novel approach which combines the revised feature score criterion with independent component analysis that has been developing recently to further improve the classification performance of gene expression data based on support vector machines. Two sets of gene expression data (colon tumor dataset and leukemia dataset) are examined to confirm that the proposed approach can extract a small quantity of independent components which drastically reduce the dimensionality of the original gene expression data when retaining higher recognition rate. For example, 100% cross-validation accuracy has been achieved with only extracting 2 or 3 independent components from leukemia dataset in our experiments.