Feature selection for classification based on gene expression profile

In order to achieve feature genes for classification, a method of feature selection based on gene expression profile was proposed according to the characters of gene expression data. In this method, an improved FDR was regarded as marking criterion of classification feature to remove the genes which are irrelevant to classification. A new distance composed of space distance and function distance was proposed as the criterion of comparability to remove redundant genes. Support vector machines as classifier test the classification performance of the feature genes. The experimental results showed that the method was effective on removing the genes which were irrelevant to the classification and redundant. The method selected least feature genes which can classify sample data accurately.