一种优化的基于网格的聚类算法  

Optimized Cell-based Clustering Algorithm

作　　者：刘俊岭[1] 孙焕良[2] 王大玲[3] 牛志成[1] 

机构地区：[1]沈阳建筑大学计算中心,辽宁沈阳110168 [2]沈阳建筑大学信息与控制工程学院,辽宁沈阳110168 [3]东北大学信息科学与工程学院,辽宁沈阳110004

出　　处：《小型微型计算机系统》2006年第10期1927-1930,共4页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(60573090)资助;辽宁自然科学基金项目(20052006)资助;辽宁省教育厅攻关计(05L354)资助.

摘　　要：聚类是数据挖掘领域中一个重要的研究课题.与其它算法相比,基于网格的聚类算法可以高效处理低维的海量数据.然而,由于划分的单元数与数据的维数呈指数增长,因此对于维数较高的数据集,生成的单元数过多,导致算法的效率较低.本文基于CD-Tree设计了新的基于网格的聚类算法,该算法的效率远高于传统的基于网格聚类算法的效率.此外,本文设计了一种剪枝优化策略,以提高算法的效率.实验表明,与传统的聚类算法相比,基于CD-Tree的聚类算法在数据集的大小及维度的可伸缩性方面均有显著提高.In data mining fields, clustering is an important issue. Comparing with other algorithms, the cell-based clustering algorithms can be applied to low dimensional data. However, in the cell-based algorithms, the number of ceils will increase exponentially with the dimensionality. So it is low efficient with high dimensionality due to a large number of cells. This paper proposes a new clustering algorithm based on CD-Tree, which improve largely the efficiency of the cell-based algorithm. In addition, to improve the efficiency of the algorithm further, we design the pruning strategy that prunes the non-dense cells before the clustering procedure. Extensive experiments on real and synthetic datasets also show that the algorithm has better scalability than other cell-based clustering algorithms.

关 键 词：数据挖掘 聚类分析 CD—Tree 基于网格的算法 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]