K-Means Clustering Efficient Algorithm with Initial Class Center Selection

Abstract

The algorithm herein adopts density-based method and max-min distance method to define initial clustering center to eliminate the need for defining clustering center in advance in k-means algorithm, and normalize the data set to reduce the influence of fluctuation of attribute value for each dimension of sample set on accuracy of clustering result.Besides, it obtains dissimilarity matrix and takes advantage of good global convergence ability of particle swarm optimization algorithm to improve proneness of K-means algorithm to be trapped in local optimum.The effectiveness of the algorithm was verified via experiment.However, although the algorithm herein performs well in part of small low dimensional data set, while how to effectively make cluster analysis on large high dimensional data still needs to be further researched.