一种基于类紧密度与相关度的聚类算法  

A Class Compactness and Relevance based Clustering Algorithm

作　　者：冯湘[1] 秦拯[1] 谷垒[1] 王雷[1] 

机构地区：[1]湖南大学软件学院,长沙410082

出　　处：《微计算机信息》2011年第2期178-179,111,共3页Control & Automation

摘　　要：针对聚类中不规则数据点分布的处理难题,提出了一种基于类紧密度的新聚类算法,在该算法中,首先通过随机选择一个较大的初始类数目并利用Voronoi图来进行聚类中心的选择,同时计算出聚类后的判决函数值;然后在每轮聚类过程中,将类数目指数递减,若当前轮得到的判决函数值小于上一轮的判决函数值,则在上一轮的类数目基础上进行线性递减,直到当再次得到当前轮的判决函数值小于上一轮的判决函数值时,将最终类数目与聚类结果设定为上一轮的类数目与聚类结果。实验结果表明,新算法具有良好的聚类效果。Aimed to solve difficult problems in clustering with irregularly distributed data set, a new clustering algorithm based on class compactness is proposed, in which, a random big number is selected as the original classes number, and then, the voronoi diagram is adopted to select the cluster centers. After clustering, the value of the decision function is computed. Thereafter, in each round of clustering, the number of clusters will be reduced exponentially until the current value of the decision function is less than that of the previous round. Then, on the basis of the class number of previous round, the number of clusters will be reduced linearly until the current value of the decision function is less than that of the previous round again. Till then, the final class number and clustering results are set to be the class number and clustering results of the previous round. Experimental results show that the new algorithm has good clustering performance.

关 键 词：聚类 VORONOI图 类紧密度 类相关度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]