A Novel Binary Particle Swarm Optimization Algorithm and Its Applications on Knapsack and Feature Selection Problems

Abstract

Particle swarm optimisation (PSO) is a well-known evolutionary computation technique, which has been applied to solve many optimisation problems. There are two main types of PSO, which are continuous PSO (CPSO) and binary PSO (BPSO). Since PSO is originally proposed to address continuous problems, CPSO has been studied extensively while there are only a few studies about BPSO. In a standard PSO algorithm, momentum is an important component, which preserves the swarm's diversity. However, since movements in binary search spaces and continuous search spaces are different, it is not appropriate to apply directly the momentum concept of CPSO to BPSO. This paper introduces a new momentum concept to BPSO, which leads to a novel BPSO algorithm, named SBPSO. SBPSO is compared with a recent BPSO algorithm, named PBPSO, in two well-known binary problems: knapsack and feature selection. The experimental results on knapsack datasets show that SBPSO can find better solutions than PBPSO. In feature selection problems, SBPSO can select a smaller number of features and still achieve similar or better accuracies than PBPSO and using all the original features in a comparative computation time.