Feature Gene Selection based on L1/2+2 Regularization

Abstract

Cancer is one of the great medical problems that mankind is facing today. With the help of DNA microarray, we can analyze thousands of genes simultaneously. The analysis of cancer samples with microarray technique is a hot topic in the field of bioinformatics. There are usually quite a lot genes in the microarray datasets, so it is time-consuming for us to classify samples with all these genes. For this reason, it is necessary for us to conduct feature gene selection. Regularization can serve as a method for feature selection. In this paper, we proposed a method called L1/2+2 and Fuzzy Measure Gene Selection (LFMGS). The method can be divided into two parts. Firstly, the L1/2+2 regularization is adopted to remove most of genes. Then L1/2+2 regularization and fuzzy measure are combined to obtain the sparse solution of fuzzy measures, and then a small number of genes are eliminated based on the final gene rank. Experimental results on seven datasets show the superiority of our method over the other four methods comprehensively considering accuracy, sensitivity and specificity, and the number of selected genes.