Multi-objective Particle Swarm Optimization based on Space Decomposition for Feature Selection

Feature selection is essentially a multi-objective optimization problem, and multi-objective particle swarm optimization is one of the effective methods for it. However, in most feature selection methods based on multi-objective particle swarm optimization, unbalanced selection pressure when determining the leading particle causes poor diversity of solutions. This paper proposes a multi-objective particle swarm optimization algorithm based on space decomposition for feature selection (MOPSO-SDFS) to solve above problem. The objective space of the external archive is equally divided into several subspaces, and a representative solution is selected for each subspace. The leading particle is chosen among these representative solutions by probability distribution. Experimental studies on 6 gene datasets verify that MOPSO-SDFS could achieve competitive classification accuracy on the most gene expression datasets.