Comparative study of feature selection methods on microarray data

It is difficult to apply usual statistical pattern recognition techniques directly to microarray data, because the number of genes is too large in comparison with the number of available training samples. Therefore, one needs a powerful feature selection method for microarray data. In this paper, we compare the previously published feature selection method with the sequential forward selection (SFS) method and the Fisher criterion-based feature selection method on the microarray data of hepatocellular carcinoma (http://surgery2.med.yamaguchi-u.ac.jp/research/DNAchip/). Experimental results show that our method outperforms the SFS method and the Fisher criterion-based method in terms of the recognition rate.