关键词: 聚类, 局部密度, 决策图, 残差分析

Abstract: K-means algorithm is a classical clustering algorithm based on partition. However, it is difficult for K-means to determine the number of clustering. Besides, K-means is sensitive to the initial centers of clustering. In order to solve the two defects of K-means algorithm, an improved K-means algorithm is proposed. Main work of this paper is putting forward a new method of calculating the density of the object, and using residual analysis method to automatically obtain the initial centers and number of clustering from the decision diagram. The result of experiment shows that the algorithm can get better clustering results.

Key words: clustering, local density, decision diagram, residual analysis