摘要
针对FCM(Fuzzy C-Means)算法对于初始聚类中心敏感,并只适合于发现球状类型簇的缺陷,提出采用冗余聚类中心初始化的方法降低算法对初始聚类中心的依赖,并先暂时将大簇或者延伸形状的簇分割成用多个小类表示,再利用隶属度矩阵提供的信息合并相邻的小类为大类,对FCM算法进行改进。实验结果显示改进的FCM算法能够在一定程度上识别不规则的簇,并减小FCM算法对初始聚类中心的依赖。
FCM（Fuzzy C-Means） algorithm has several defects,being sensitive to the initial cluster centers,only being applied to the type found in globular clusters.There are several methods used for reducing the algorithm dependence on the initial cluster centers,expressing big cluster or extension of shape used several small clusters.At last,this paper merges adjacent small cluster into a big cluster,using the information provided by partition matrix.Experiment result demonstrates that the improved FCM algorithm can distinguish irregular cluster to a certain extent,decrease the dependence on the initial cluster centers.
出处
《计算机工程与应用》
CSCD
北大核心
2009年第24期122-124,共3页
Computer Engineering and Applications
基金
中国博士后科学基金No20070420711
重庆市科委自然科学基金计划资助项目No2007BB2372~~
关键词
模糊聚类
模糊C均值算法
隶属度矩阵
fuzzy clustering
Fuzzy C-Means algorithm（FCM）
partition matrix
