摘要
为估计数据集的聚类数目及获得较好的聚类性能,提出了一种基于灰关联测度的分裂式层次聚类算法.该算法用灰关联测度衡量数据对象之间的相似程度,以基于密度扩展的方式自顶向下分裂成不同层次的数据集划分;然后,根据灰关联测度定义聚类有效性指标;最后将有效性指标曲线极值点对应的聚类划分用于估计最佳聚类数目.实际数据和合成数据集的实验表明,与FCM聚类相比,该算法的聚类正确率平均提高3.7%,并且能够识别任意形状的簇.
To estimate cluster number and achieve a better clustering performance,a divisive hierarchical clustering algorithm based on grey relational measure was proposed.In this algorithm,the grey relational measure is used to measure the degree of similarity between data sets.On the basis of the way of density-based extension,the algorithm divisively generates hierarchical partitions of data set.And then the clustering validity index is defined based on the grey relational measure.The partitions corresponding to the extremum of the validity index curve are used to estimate the number of clusters finally.Computer simulation on real and synthesis data sets shows that compared with the FCM(fuzzy C-means) algorithm,the proposed algorithm has a 3.7% improvement in average clustering correct rate and is good for arbitrary-shaped clusters.
出处
《西南交通大学学报》
EI
CSCD
北大核心
2010年第2期296-301,共6页
Journal of Southwest Jiaotong University
基金
国家自然科学基金资助项目(60971103)
关键词
灰关联测度
聚类分析
层次聚类
聚类有效性指标
grey relational measure
clustering analysis
hierarchical clustering
clustering validity index