Feature Selection Method with Proportionate Fitness Based Binary Particle Swarm Optimization

Particle swarm optimization(PSO) has been applied on feature selection with many improved results. Traditional PSO methods have some drawbacks when dealing with binary space, which may have negative effects on the selection result. In this paper, an algorithm based on fitness proportionate selection binary particle swarm optimization(FPSBPSO) will be discussed in detail aiming to overcome the problems of traditional PSO methods. FPSBPSO will be utilized in the feature subset selection domain. The performance of feature selection will be compared in a benchmark dataset, and experimental results prove that the FPSBPSO-based feature selection methods can avoid premature convergence and improve the classification accuracy.