Gene Selection Using Neighborhood Rough Set from Gene Expression Profiles

Although adopting feature reduction in classic rough set theory to select informative genes is an effective method, its classification accuracy rate is usually not higher compared with other tumor-related gene selection and tumor classification approaches; for gene expression values must be discretized before gene reduction, which leads to information loss in tumor classification. Therefore, the neighborhood rough set model proposed by Hu Qing-Hua is introduced to tumor classification, which omits the discretization procedure, so no information loss occurs before gene reduction. Experiments on two well-known tumor datasets show that gene selection using neighborhood rough set model obviously outperforms using classic rough set theory and experiment results also prove that the most of the selected gene subset not only has higher accuracy rate but also are related to tumor.