摘要
在众多聚类算法中,基于划分的模糊聚类算法是模式识别中最常用的算法类型之一.至今,文献中仍不断有关于基于划分的模糊聚类算法的研究成果出现.为了能更为系统和深入地了解这些聚类算法及其性质,本文从改变度量方式、改变约束条件、在目标函数中引入熵以及考虑对聚类中心进行约束等几个方面,对在C-均值算法的基础上得到的基于划分的模糊聚类算法作了综述和评价,对各典型算法的优缺点进行了实验比较分析.指出标准FCM算法被广泛应用的原因之一是它对数据的比例变化具有鲁棒性,而其他类似的算法对这种比例变化却很敏感,并以极大熵方法为例进行了比较实验.最后总结了基于划分的模糊聚类算法普遍存在的问题及其发展前景.
Fuzzy partitional clustering algorithms are widely used in pattern recognition field. Until now, more and more research results on them have been developed in the literature. In order to study these algorithms systematically and deeply, they are reviewed in this paper based on c-means algorithm, from metrics, entropy, and constraints on membership function or cluster centers. Moreover, the advantages and disadvantages of the typical fuzzy partitional algorithms are discussed. It is pointed out that the standard FCM algorithm is robust to the scaling transformation of dataset, while others are sensitive to such transformation. Such conclusion is experimentally verified when implementing the standard FCM and the maximum entropy clustering algorithm. Finally, the problems existing in these algorithms and the prospects of the fuzzy partitional algorithms are discussed.
出处
《软件学报》
EI
CSCD
北大核心
2004年第6期858-868,共11页
Journal of Software
基金
国家自然科学基金
教育部科学技术研究重点项目~~
关键词
划分聚类
C均值
权重指数
熵
隶属度函数
partitional clustering
C-means
weighting exponent
entropy
membership function
