Multi-objective simplified swarm optimization with weighting scheme for gene selection

Abstract

Gene selection can be regarded as a multi-objective problem which involves both minimizing the size of a gene subset and maximizing the prediction performance. This work proposes a hybrid filter/wrapper method for gene selection based on multi-objective optimization. In this method, an emerging aggregate filter method is adopted as a filter with which to choose the most informative genes; in addition, a multi-objective simplified swarm optimization (MOSSO) is proposed and integrated with a support vector machine as a wrapper to seek an optimal gene subset from the selected genes. Unlike most current multi-objective based methods employed to handle gene selection problems, the proposed MOSSO uses a weighting scheme to guide the search towards the interesting regions as defined by the preference, which means that not all Pareto optimal solutions are generated, but only the ones gene selection prefers. The proposed method is validated using ten gene expression datasets, and the corresponding results are compared with those obtained with existing works. Statistical analysis indicates that the proposed method is highly competitive and, can be considered a promising alternative for dealing with gene selection problems.