A parameter-free barebones particle swarm algorithm for unsupervised pattern classification

Abstract

This paper introduces an efficient algorithm for unsupervised clustering that is based on barebones Particle Swarm (BB). The proposed algorithm introduces significant enhancement to the Particle Swarm Optimization (PSO) by eliminating the parameters tuning. The Algorithm aims at finding the centroids of predefined number of clusters where each centroid attracts similar patterns. This research tests and investigates the application of the proposed algorithm to the problem of unsupervised pattern classification by applying the algorithm to segmentations of different images. Experimental results show that the the proposed BB-based algorithm outperforms other state-of-the-art clustering algorithms on all the different levels of comparison. The impact of eliminating the parameters tuning is evident on the performance of the algorithm. In addition, the influence of different values for the swarm size of BB on performance is also illustrated.