Gene Selection and Classification Method Based on SNR and Multi-loops BPSO

Abstract

For gene expression data with a massive amount of redundant data and noise, gene selection methods based on binary particle swarm optimization algorithm (BPSO) is an important method to improve classification performance. However, most BPSO-based methods can only handle finite sets, resulting in more selected genes and lower classification accuracy. This paper proposes a hybrid method SNR-MBPSO that utilizes signal-to-noise ratio (SNR) combined with multi-loops BPSO and various classifiers. It is important to point out that the multi-loops BPSO structure in this paper is first proposed and used to solve gene selection problems. And the multi-loops BPSO shows a remarkable improvement in the selection when it is combined with SNR. What is more, the traditional BPSO is modified by using adaptive weights and an improved bit-value changing strategy. To verify the performance of the proposed method, the SNR-MBPSO is compared with the other seven recently published algorithms in the literature. Experimental results based on nine publicly available gene expression datasets have shown that the proposed method significantly outperforms the state-of-the-art methods in terms of classification accuracy and the number of key genes.