Improving the prediction accuracy in classification using the combined data sets by ranks of gene expressions