Fast and Effective Classification using Parallel and Multi-start PSO

Abstract

PSO being a swarm based algorithm, can efficiently lend itself to operate on huge data. This article presents a technique that performs classification using PSO. An initial discussion is carried out describing PSO as a classifier. Three variants of PSO are proposed here; the first variant hybridizes PSO using Simulated Annealing and the next two variants parallelizes PSO. The two parallel variants of PSO are; Parallel PSO and Multistart PSO. Parallel PSO operates by parallelizing the operation of each of the particles and Multistart PSO runs several normal versions of PSO embedded with Simulated Annealing in parallel. The multi-start version is implemented to eliminate the problem of local optima. Experiments were conducted to identify the scalability and efficiency of PSO and its variants on huge and imbalanced data.