Kmeans聚类分析算法中一个新的确定聚类个数有效性的指标_李双虎.pdf

20 4 　 　 　 　 　 　 　 Vol.20 No.4 2003 11 Journal of the Hebei A ademy of S ien es Nov.2003 :1001-9383(2003)04 -0199-04 K-means 1 2 李双虎, 王铁洪 (1., 050081; 2., 050081) 【 　】　K-means 算法是聚类分析中使用最为广泛的算法之 一。然而, 该算法通常受到初 [ 1] 聚类条件的影响。关于这 个问题的详细讨论可参看文献 。该算法的另 一个不足之处是, 聚类数目K 必须作为参数由用户提供。笔者提出了 一个新的有关聚类有效性的度量指标和优 化的K-means 算法。它能自动确定最佳聚类 个数。 【】　聚类分析;K-means 算法;有效性度量;指标 【】　TP 301.6 　　　【】　A New validity index for determining the number of clusters in K-means clustering 1 2 LI Shuang-hu , WA NG Tie-hong (1 .Institute of App lied Ma th ..Hebei Academy of Sciences, Sh ij iaz huang 050081, China; 2 .Institute of A utomation , Hebei Academy of Sciences, Shij iaz huang 050081, China ) Abstract 　K-Means Clustering Algorithm is one of the most popular methods in luster analysis. How ever, it is well know n that K-means algorithm suffers from initial starting onditions effe ts(initial lustering and instan e order effe ts).For more detailed dis ussion on initialization methods, see literature [1] .Another w eakness of k-means algorithm is that the number of lusters, , must be supplied as a parameter.In this paper, a new validity measure for k-means lustering is presented to allow the number of lusters to be determined automati ally . Keywords 　Cluster analysis;K-Means Algorithm ;Validity measure;Index (), “” “”, “”。: (1)K 。, 。 (2), “”。 1 　K-means K-means : n 　　　　MinimizeJ (X , U, V)=