节点优先级导向的聚类算法  

Node priority guided clustering algorithm

作　　者：宗瑜[1,2] 徐贯东[2] 张彦春[2] 李明楚[1] 

机构地区：[1]大连理工大学软件学院,辽宁大连116621 [2]澳大利亚维多利亚大学信息应用中心

出　　处：《控制与决策》2011年第6期879-882,887,共5页Control and Decision

基　　金：国家自然科学基金项目(60503003);国家973计划项目(2007CB714205);安徽省教育厅重点项目(KJ2009A54)

摘　　要：基于密度的聚类算法具有挖掘任意形状聚类结果和处理"噪声"数据等优势,同时也存在无法处理高维和密度分布不均匀数据的缺陷;鉴于此,给出了节点优先级导向的聚类算法.首先建立数据集的有向K邻居图;然后用K-最近邻核密度估计方法获得数据对象的局部信息,并在图中迭代地传播,以产生数据对象的优先级;最后以该优先级为导向从图中搜索聚类结果,实验结果表明,该算法适合处理高维、密度分布不均匀的数据.Density-based clustering algorithms have the advantages of clustering with arbitrary shapes and handling noise data, but cannot deal with unsymmetrical density distribution and high dimensionality dataset. Therefore, a node priority guided clustering algorithm（NPGC） is proposed. A direct K neighbor graph of dataset is set up based on KNN neighbor method. Then the local information of each node in graph is captured by using KNN kernel density estimate method, and the node priority is calculated by passing the local information through graph. Finally, a depth-first search on graph is applied to find out the clustering results based on the local kernel degree. Experiment results show that NPGC has the ability to deal with unsymmetrical density distribution and high dimensionality dataset.

关 键 词：密度聚类 K-最近邻核密度 节点优先级 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]