一种基于平行坐标度量模型的聚类算法及其应用  

A clustering algorithm for parallel coordinates-based measure model and its applications

作　　者：胡俊[1,2] 黄厚宽[1] 高芳[1] 

机构地区：[1]北京交通大学计算机与信息技术学院,北京100044 [2]新疆石河子大学信息科学与技术学院,石河子832003

出　　处：《南京大学学报（自然科学版）》2009年第5期645-655,共11页Journal of Nanjing University（Natural Science）

基　　金：国家"973"重点基础研究发展规划(2007CB307100;2007CB307106);国家自然科学基金(60673089);北京交通大学基金(2004SM009)

摘　　要：本文说明了数据挖掘中可视化技术应用的特点与方法,给出了数据挖掘中可视对象与参数的确定及算法分解的方法,并给出基于平行坐标技术的聚类算法的可视化方法与平行坐标的度量模型,以及在K-means算法上的应用方法.结果表明这种方法对于数据及聚类算法K-means的数据挖掘过程的可视化表示是有效的.To apply visualization to data mining, or to establish visible data mining method is a cross research subject about visualization and data mining. This type of research requires to be established above reasonable acknowledge basement. On one hand, it requires to analyse the theory and technology basement of this method; on the other hand, it also requires to consider the visualization character of the property of the data mining subject and the observer＇s awareness of visualization character. Two aspects need to be considered during applying visualization to cluster analysis method： one is the separability of the cluster algorithm process, that is, to split the process of the cluster algorithm doesn＇t affect the result of the cluster; the other is determining the key factors in the cluster algorithm and measuring standard, and then finding out their influences on the result of the cluster. The K-means algorithm chooses the expected number of the cluster centers in the dataset, and alters the centers, to find the minimum variance in the whole cluster, so as to find out the cluster and cluster center of the dataset. There are several key parameters in the K-means algorithm, and they have key effect on the result of the data mining, so it is able to determine the object of the visualization basing on the feature of the K-means algorithm and the parameters. To explain the visualization result of multiple-dimension data in a better way, the process of the visualization＇s objects should match with the process of the data objects, so it＇s able to draw necessary measureing index into visualization application. To draw into proper measure index in quantification contributes to improving the visualization technology, designing applicable measuring index, and establishing functional evaluation model. The application of the data mining algorithm based on visible measuring index provides a visible data mining method. In the process of the clustering, visualization technology contributes to finding out the clustering

关 键 词：数据挖掘 可视化 聚类 平行坐标法 度量 

分 类 号：TP311.132[自动化与计算机技术—计算机软件与理论]