摘要
在分析影响C均值法聚类效果的两个主要因素的基础上,将紧致性的概念与基于密度的初始聚类中心的选取方法和类的初始划分方法相结合,提出了一种改进划分初始类的方法。此方法用"距离试探法"来确定一个合适的d0值,以此为基础进行马氏距离测度下类的初始划分,再以临界函数作为紧致性的判断依据,修改半径d0,得到新的聚类中心,从而提高了聚类的效果。
After the analysis of two main factors on the effcet of C - means algorithm, based on the density about selecting initial cluster centre and initial allocation, a new method to improve the way of division initial allocation was proposed. By ＂distance test method＂, an appropriate value d0 based Mahalanobis Distance on division initial allocation was determined. Taking the critical function as a judgment to change the vahle d0, a new initial cluster centre was gained to imrove the clustering effect.
出处
《武汉理工大学学报（信息与管理工程版）》
CAS
2008年第2期209-211,214,共4页
Journal of Wuhan University of Technology：Information & Management Engineering
关键词
C均值法
初始聚类中心
紧致性
临界函数
C- means algorithm
initial cluster centre
compactness
critical function
