基于划分和层次的混合动态聚类算法  

Hybrid dynamic clustering algorithm based on partition and hierarchical clustering

作　　者：郝洪星[1] 朱玉全[1] 陈耿[2] 李米娜[1] 

机构地区：[1]江苏大学计算机科学与通信工程学院,江苏镇江212013 [2]南京审计学院信息科学学院,南京211815

出　　处：《计算机应用研究》2011年第1期51-53,共3页Application Research of Computers

基　　金：江苏省"青蓝工程";江苏省六大人才高峰项目(07-E-025);江苏省高校自然科学重大基金研究项目(08KJA520001);国家中小企业创新基金资助项目(09C26213203797);国家自然科学基金资助项目(70971067)

摘　　要：针对划分聚类对初始值较为敏感以及层次聚类时间复杂度高等缺陷,提出了一种基于划分和层次的混合动态聚类算法HDC-PH。该算法首先使用划分聚类快速生成一定数量的子簇,然后以整体相似度的聚类质量评价标准来动态改变聚类数目,同时给出了聚类过程中孤立点的剔除方法。实验结果表明,HDC-PH算法的性能明显优于划分和层次算法,提高了聚类质量,并获得了更自然的聚类结果。For resolving the problem that partition clustering algorithms are sensitive to initial value and the time complexity of hierarchical clustering algorithms is high, this paper proposed a new hybrid dynamic clustering algorithm called HDC-PH, which was based on partition and hierarchical method. At first, HDC-PH partitioned the input data set into a number of subclusters, and then in a hierarchical manner, continuously merged the suhclusters dynamically by using overall similarity to evaluate the cluster quality. At the same time, described a method to explain how to eliminate outlier during clustering. Experimental results show that HDC-PH is better than partition and hierarchical algorithms and get the better clustering results.

关 键 词：K-MEANS CURE 混合聚类 孤立点 整体相似度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]