.

Microarray analysis allows scientists to screen thousands of genes and determine the active, hyperactive or silent genes in normal or cancerous tissues. Hence, analytical methods should be developed to distinguish between cancer tissues of gene expression over normal tissues or other cancer tissues type. In this paper, gene selection of small subset of gene using component-wise two sample  statistic is performed on gene expression data. Moreover, we seek to find the effect of gene selection technique on support vector machine (SVM) for classification of cancerous gene level and micro-array data in general. The gene expression datasets for this paper are colon cancer data, leukaemia data and prostate cancer data. Findings from this study showed that discrimination between cancerous patients and non-cancerous patients using SVM when gene selection based on  test is employed tends to be much better than when all the genes are used in terms of probabilities of correct classification. Also, the statistical test of location can only be performed on selected genes rather than full datasets because of sparsity of gene expression data. Using the selected genes, a statistical test of location parameters in  is examined for different classes in gene expression data.  
Keywords: Gene expression,  statistic, SVM, test of location parameter, gene selection.