计算机工程与应用 ›› 2008, Vol. 44 ›› Issue (22): 65-67.DOI: 10.3778/j.issn.1002-8331.2008.22.019

• 理论研究 • 上一篇    下一篇

利用FCM求解最佳聚类数的算法

张姣玲   

  1. 广东技术师范学院 计算机科学学院,广州 510665
  • 收稿日期:2007-10-11 修回日期:2008-01-24 出版日期:2008-07-11 发布日期:2008-07-11
  • 通讯作者: 张姣玲

FCM-based algorithms for determining optimal number of clusters

ZHANG Jiao-ling   

  1. Department of Computer Science,Guangdong Polytechnic Normal University,Guangzhou 510665,China
  • Received:2007-10-11 Revised:2008-01-24 Online:2008-07-11 Published:2008-07-11
  • Contact: ZHANG Jiao-ling

摘要: 利用FCM求解最佳聚类数的算法中,每次应用FCM算法都要重新初始化类中心,而FCM算法对初始类中心敏感,这样使得利用FCM求解最佳聚类数的算法很不稳定。对该算法进行了改进,提出了一个合并函数,使得(c-1)类的类中心依赖于类的类中心。仿真实验表明:新的算法稳定性好,且运算速度明显比旧的算法要快。

关键词: 聚类, 模糊C-均值, 有效性函数

Abstract: In the conventional FCM-based algorithm for determining the optimal number of clusters,the authors use random initialization at the beginning of each clustering phase.By doing so,the conventional FCM-based algorithm is unstable because different cluster centers lead different clustering results in FCM algorithm.In this paper,a new algorithm is proposed to improve the conventional algorithm by reducing the randomness in the initialization of cluster centers at the beginning of each clustering phase.Experimental results indicate that the new algorithm is stable and its speed is faster than the conventional algorithm.

Key words: clustering, fuzzy C-means, validity index