Gene selection based on fuzzy measure with L1 regularization

Gene selection is very important for cancer classification in genomic data analysis. We need deal with high-dimensional gene space and few samples. There have been many methods with L1 Regulation to reduce genes number using sparsity. These shifted genes are considered as key genes for disease. But the epistasis means some genes maybe cover or affect other genes. Fuzzy measure can describe the interaction in genes very well. It is related to the power set of gene set, so the computing complexity is very tremendous for huge gene space. In this article, we proposed one new gene selection method which is based on fuzzy measure with sparse solutions using L1 regulation, FMSS for short. Fuzzy integral is combined with fuzzy measure to construct linear equations, which is a kind of fusion tool to solve nonlinear problems. A group of gene combinations can be obtained corresponding to the fewest nonzero fuzzy measure values. Meanwhile, the important gene or genes can be selected according to frequency of appearance in gene subsets. The new method is applied to one cancer dataset for testifying the performance. Experimental results show that the proposed method has highly competitive performance compared with several state-of-the art methods. FMSS can output the highest accuracy and important gene subset.