This paper presents an evolutionary particle swarm optimization (PSO) learning-based method to optimally cluster N data points into K clusters. The hybrid PSO and K-means algorithm with a novel alternative metric, called Alternative KPSO-clustering (AKPSO), is developed to automatically detect the cluster centers of geometrical structure data sets. The alternative metric is known has more robust ability than the common-used Euclidean norm. In AKPSO algorithm, the special alternative metric is considered to improve the traditional K-means clustering algorithm to deal with various structure data sets. For testing the performance of the proposed method, this paper will show the experience results by using several artificial and real data sets. Simulation results compared with some well-known clustering methods demonstrate the robustness and efficiency of the novel AKPSO method.
淡江理工學刊=Tamkang journal of science and engineering 8(2), pp.165-174