基于近邻传播算法的最佳聚类数确定方法比较研究  

Comparative Study on Method for Determining Optimal Number of Clusters Based on Affinity Propagation Clustering

作　　者：周世兵[1] 徐振源[1,2] 唐旭清[2] 

机构地区：[1]江南大学信息工程学院,无锡214122 [2]江南大学理学院,无锡214122

出　　处：《计算机科学》2011年第2期225-228,共4页Computer Science

基　　金：国家863计划项目(2007AA1Z158);国家自然科学基金(60703106)资助。

摘　　要：在聚类分析中,决定聚类质量的关键是确定最佳聚类数。提出采用聚类效果较好的近邻传播聚类算法对样本进行聚类,运用6种聚类有效性指标分别对聚类结果进行有效性分析,以确定最佳聚类数。具体分析了这些有效性指标,并改进了IGP指标确定最佳聚类数的方法。针对8个数据集,通过实验比较这些指标的性能。分析和实验结果表明,基于近邻传播聚类算法,IGP指标确定最佳聚类数的性能最好。It is crucial to determine optimal number of clusters for the quality of clustering in cluster analysis.Based on Affinity Propagation clustering algorithm,a method for determining optimal number of clusters was proposed to analyze the clustering validity and determine optimal number of clusters by using six clustering validity index.These clustering validity indexes were analyzed concretely and the method of using IGP index to determine optimal number of clusters was improved.In connection with eight datasets,the performances of these indexes were compared by simulation experiments.The results of analysis and experiments show that IGP index is the best to determine optimal number of clusters based on Affinity Propagation clustering.

关 键 词：近邻传播 聚类数 聚类有效性指标 聚类分析 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]