摘要：The initial clustering centers of traditional bisecting K-means algorithm are randomly selected andthe k value of traditional bisecting K-means algorithm could not determine beforehand. This paper proposesa improve bisecting K-means algorithm based on automatically determining K value and the optimization ofthe cluster center. Firstly, the initial cluster centers are selected by using the point density and the distancefunction; Secondly, automatically determining K value is proposed by using Intra cluster similarity and intercluster difference. the experiment results on UCI database show that the algorithm can effectively avoid theinfluence of noise points and outliers, and improve the accuracy and stability of clustering results.
