一种最小生成树聚类算法  

Minimum Spanning Tree Clustering Algorithm

作　　者：王小乐[1] 刘青宝[1] 陆昌辉[1] 侯东风[1] 

机构地区：[1]国防科学技术大学信息系统与管理学院,湖南长沙410073

出　　处：《小型微型计算机系统》2009年第5期877-882,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(70771110)资助

摘　　要：现有的聚类算法都不能在输入较少参数的情况下得到任意形状任意密度的类.提出一种最小生成树的聚类算法,该算法不但能解决上述问题,还能处理高维数据,发现异常点,且具有扩展性.针对该算法提出一个目标函数,该函数根据对象的类属情况和相似度统计信息来判别聚类效果的质量.最后,通过实验验证了该算法的聚类质量很好,目标函数具有良好的收敛性.The existing clustering algorithm can not discover clusters with arbitrary shape and multi-density using few parameters. In this paper we present a new clustering algorithm named MSTClust which is based on minimum spanning tree. The MSTCIust can discover clusters with arbitrary shape and multi-density, can dispose multidimensional data, can detect outer point and have a good expansibility. In allusion to MSTClust we propose an objective function which refers to statistical Information of the weight of edges in minimum spanning tree. Finally the experimental result showed the effectiveness and efficiency of MSTClust and proved that the objective function have good astringency.

关 键 词：聚类 相似度量 最小生成树 聚类目标函数 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]