A Recursive Information Gene Selection Using Improved Laplacian Maximum Margin Criterion

Gene selection is an important research topic in pattern recognition and tumor classification. Numerous methods have been proposed, Maximum Margin Criterion (MMC) is one of the famous methods have been proposed to solve the small size samples problem. But, the MMC only considers the global structure of samples. In this article, a novel recursive gene selection criterion named Laplacian Maximum Margin Criterion Recursive Feature Elimination (LMMC-RFE) is proposed to address this issue. The Laplacian within-class scatter matrix and Laplacian between-class scatter matrix can be formulated by the use of LoG weighted matrix to capture the scatter information. The neighboring structure of the withinclass samples is preserving while the samples from different classes are mapped far from each other. The successful application to the Colon dataset, Prostate dataset and Leukemia dataset suggests this proposed method is more efficient and competitive in gene selection.