一种最小生成树聚类算法概论.pdfVIP

2009 5 5 Journal o f Chinese Com puter Sy stem s V o l30 N o. 5 2009 王小乐, 刘青宝, 陆昌辉, 侯东风 , 4 10073) E-m ail: w x -l 54 @ 163. com : 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 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. : ; ; ; : T P39 1 : A : 1000-1220 2009) 05-0877-06 M inimum Spanning Tree Clustering A lgorithm W ANG X iao- le, L IU Q ing-bao, LU Chang-hui, HO U D ong- feng C ollege o f In form at ion Sy stem and M ana gem en t, N a tiona l U niversity of D efen se T echno logy, C hangsha 4 10073, C h ina ) Abstract: T he ex isting c lu stering a lgo r ithm can no t discov er clu sters w ith arb itra ry shape and m u lti-density u sing few param eters. In th is paper w e present a new c lu ste ring a lg o rithm nam ed M STC lu st w hich is ba sed o n m in mi um spann ing tree. T he M STC lu st can dis- cov er c lu ste rs w ith a rbitra ry shape and m u lt-i density, can d ispo se m ultid mi ensiona l data, can de tect o uter po in t and hav e a go od ex- pansib ility. In a llu sion to M STC lu st w e pro po se an o bjectiv e function w hich ref ers to statistical Info rm a tion o f the w e igh t o f edge s in m in mi um spanning tree. F ina lly the expermi en tal resu lt show ed the eff ectiv ene ss and eff ic iency o f M STC lu st and pro ved tha t the o b- jectiv e function hav e goo d astringency. K ey words: c lu stering; smi ilar ity m easure; m in mi um spann ing tree; c lu ste ring o bjectiv e function 1 . M ST , M ST, , , , , . [ 1] , 1. , ) , , , . , ; , ; , , . : [ 2] [ 3] [ 4] , , ,