一种基于几何中心的模糊c均值聚类方法  

A fuzzy c-means clustering algorithm based on geometrical center

作　　者：张元武[1] 王丽珍[1] 孔云[1] 

机构地区：[1]云南大学信息学院,云南昆明650091

出　　处：《云南大学学报（自然科学版）》2009年第S1期189-193,共5页Journal of Yunnan University(Natural Sciences Edition)

基　　金：国家自然科学基金资助项目(60463004)

摘　　要：对模糊c均值聚类算法容易陷入局部极值点和鞍点而得不到全局最优模糊划分进行研究,提出了一种新颖的基于几何中心的模糊c均值聚类(GCFCM)方法,利用网格和密度找到数据集的近似聚类中心.实验结果表明:该算法与经典FCM聚类算法相比收敛速度更快,迭代步数更少,解决了容易陷入局部极值点和鞍点的问题.It is discussed that Fuzzy c-means clustering algorithm(FCM) is apt to forming partial extreme point and saddle point,so it fail to get global optimization of fuzzy partition. Following that it is proposed a fuzzy c-means clustering algorithm based on geometrical center(GCFCM),which applies grid and density to discover the approximate center of sample. That the algorithm is faster in convergence speed,less steps in iteration,comparing to classical FCM is shown by the experimental.And it avoids classical FCM's weakness that it is apt to forming partial extreme points and saddle point.

关 键 词：模糊c均值聚类(FCM)算法 网格 密度 初始化聚类中心 二分法 几何中心 

分 类 号：O159[理学—数学]