Set-based Particle Swarm Optimization for Data Clustering: Comparison and Analysis of Control Parameters

Abstract

Data clustering is a highly studied field of data science and computational intelligence. Population-based algorithms such as particle swarm optimization (PSO) have shown to be effective at data clustering. Set-based particle swarm optimization (SBPSO) is a generic set-based PSO variant that has shown promise in clustering stationary and non-stationary data. In this paper, SBPSO is used to cluster fifteen datasets with diverse characteristics. The clustering ability of SBPSO is compared in depth to the performance of six other tuned clustering algorithms. A sensitivity analysis of the SBPSO control parameters is performed to determine the effect that variation in these control parameters have on swarm diversity and other measures. SBPSO ranked third from among the algorithms evaluated and proved a viable clustering algorithm. A trade-off between swarm diversity and clustering ability was discovered, and the control parameters that control this trade-off were determined.