一种基于网格密度与空间划分树的聚类算法  

A clustering algorithm based on the grid-density and the spatial partition tree

作　　者：曾东海[1] 米红[2] 刘力丰[1] 

机构地区：[1]厦门大学信息科学与技术学院模式识别与智能系统研究所,厦门361005 [2]浙江大学公共管理学院,杭州310027

出　　处：《系统工程理论与实践》2008年第7期125-131,137,共8页Systems Engineering-Theory & Practice

基　　金：厦门大学“985”工程二期项目“国防信息化安全智能创新平台”资助

摘　　要：基于密度的聚类是聚类分析中的一种,其主要优点是发现任意形状的聚类和对噪音数据不敏感.文章提出了一种新的基于网格密度和空间划分树的CGDSPT(Clustering based on Grid-Density and Spatial Partition Tree)聚类算法.其创新点在于,将数据空间划分成多个体积相等的单元格,然后基于单元格定义了密度、簇等概念,对单元格建立了一种基于空间划分的空间索引结构(空间划分树)来对数据进行聚类.CGDSPT算法保持了基于密度的聚类算法的上述优点,而且CGDSPT算法具有线性的时间复杂性,因此CGDSPT算法适合对大规模数据的挖掘.理论分析和实验结果也证明了CGDSPT算法的优点.The density-based clustering algorithm is a sort of clustering analysis, its main merit is to discover arbitrary shape cluster and is insensitive to the noise data. This paper proposed a new clustering algorithm based on the grid density and the spatial partition tree CGDSPT. It is able to cluster data through dividing the data space into several unit cells. Some concepts, for example： the density, the bunch and so on, are defined on the unit cell. Then we established a spatial index structure for spatial division. The CGDSFT inherits the merit of the density-based clustering algorithm, moreover CGDSFT has the linear time-complexity, therefore it suits to the large-scale data mining. The theoretical analysis and the experimental result have also proven the merit of CGDSPT.

关 键 词：聚类 网格密度 空间划分树 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]