关键词: K-means算法, 聚类, 对称距离, 方向约束

Abstract: K-means is a well studied and widely used clustering algorithm in data mining. There are many clustering algorithms evolved from K-means. For example, the symmetry-based version of the K-means algorithm using the point symmetry distance as the similarity measure is proposed at recent years. In this paper, a new clustering algorithm based on point symmetry distance clustering algorithm is proposed. The direction constraint is put forward after studying the pro-perties of symmetry to enhance the description of symmetric distance and improve the accuracy of clustering. For the fact that symmetry is the relationship between two points, the strategy of convergence is modified to use the midpoint of the symmetry pair to calculate the cluster centers. The convergence performance of clustering is improved. By numerical simulation it shows that the proposed algorithm reaches a more accurate result with the same computational complexity as the existing one.

Key words: K-means algorithm, clustering, symmetry distance, direction constraint