Set-based Particle Swarm Optimization for Data Clustering

Abstract

Computational intelligence approaches to data clustering have been successful in producing compact and well-separated clusters. In particular, particle swarm optimization (PSO) is deemed an effective approach to data clustering. This paper develops and evaluates a discrete-valued variation of PSO, namely the set-based PSO (SBPSO) algorithm, to cluster data. The SBPSO algorithm is evaluated on six standard data sets and nine artificially generated data sets. The clustering results of the SBPSO algorithm is compared to the performance of established clustering algorithms and a PSO clustering algorithm. It is concluded that the results of the SBPSO algorithm varies with the data set characteristics. Nonetheless, the SBPSO is deemed a successful approach to clustering data.