An efficient hybrid approach based on PSO, ABC and k-means for cluster analysis

Abstract

Particle Swarm Optimization (PSO) algorithm is one of the typical example of Swarm Intelligence (SI) algorithm. This article addresses such problems of PSO algorithm as random initial position of each particle, unsmooth speed weight change, and poor search ability, and proposes an optimization algorithm—hybrid PSO (HPSO) algorithm to solve these problems. This algorithm makes comprehensive improvements to the PSO clustering algorithm by using the K-means clustering algorithm to generate initial clustering centers, adopting a negative exponential function model to update the weight of velocity when constructing the “position-velocity” model, and introducing the “search restriction” mechanism, and the “fly-back” mechanism and auxiliary search methods such as the single point crossover operator in the Artificial Bee Colony (ABC) algorithm. Furthermore, experimental results were analyzed and verified. The experiment compares HPSO algorithm with K-Means algorithm, PSO algorithm, and other two typical improved algorithms from the literature on six of the UCI standard clustering test data sets. The results indicate that HPSO algorithm has good performance in stability, clustering effectiveness, robustness and global search ability.