A fuzzy extension of the silhouette width criterion for cluster analysis
A fuzzy extension of the silhouette width criterion for cluster analysis
复制标题
DOI：
10.1016/j.fss.2006.07.006
复制
发表时间：
2006-11-01
影响因子：
3.9
通讯作者：
Hruschka, E. R.
中科院分区：
文献类型：
--
作者：
Campello, R. J. G. B.;Hruschka, E. R.
本文提出了一种新的聚类有效性测度，作为模糊聚类分析中辅助决策的附加准则。这种度量称为模糊轮廓，是对平均轮廓宽度标准的模糊情况的概括，最初设想用于评估清晰（非模糊）数据分区。模糊轮廓是更有吸引力的模糊聚类分析的上下文中，因为它明确使用的聚类算法提供的模糊划分矩阵比其清晰的对应。此外，它已被设计为提高性能的原始轮廓标准检测区域具有较高的数据密度时，数据集涉及重叠的集群。模糊轮廓的性能进行评估和比较，五个著名的聚类有效性措施。六个数据集被用来说明不同的情况下，建议的模糊轮廓执行类似或优于这些其他标准，从而成为合格的加入池的措施一起使用的模糊聚类分析。(C)2006 Elsevier B.V.保留所有权利。
The present paper proposes a new cluster validity measure as an additional criterion to help the decision making process in fuzzy cluster analysis. This measure, named Fuzzy Silhouette, is a generalization to the fuzzy case of the Average Silhouette Width Criterion, originally conceived to assess crisp (non-fuzzy) data partitions. The Fuzzy Silhouette is more appealing than its crisp counterpart in the context of fuzzy cluster analysis since it makes explicit use of the fuzzy partition matrix provided by the clustering algorithm. In addition, it has been designed to improve performance of the original silhouette criterion in detecting regions with higher data density when the data set involves overlapping clusters. The performance of the Fuzzy Silhouette is evaluated and compared to that of five well-known cluster validity measures. Six data sets are used to illustrate different scenarios in which the proposed Fuzzy Silhouette performs similar to or better than these other criteria, thus becoming eligible to join a pool of measures to be used all together in fuzzy cluster analysis. (C) 2006 Elsevier B.V. All rights reserved.