关键词: 特征选择，拆分，主表，子表，粗糙集，基因选择

Abstract: Efficient gene selection is a key issue for classifying microarray gene expression data, since the data typically consist of a huge number of genes and a few dozens of samples. Rough set theory is an efficient tool for further reducing redundancy. However, when handling numerous genes, most existing methods based on rough set theory gain worse performance. A gene selection method based on decomposition was presented. The idea of decomposition is to break a complex task down into a master-task and several sub-tasks that are simpler, more manageable and more solvable by using existing induction methods, then joining them together to solve the original task. To evaluate the performance of the proposed approach, we applied it to four benchmark gene expression data sets and compared our results with those obtained by conventional methods. Experimental results illustrate that our algorithm improve computational efficiency significantly while keeping classification accuracy.

Key words: Feature selection, Decomposition, Master-table, Sub-table, Rough set theory, Uene selection