一种新的基于粗糙集的leader聚类算法  

A Novel Rough-based Leader Clustering Algorithm

作　　者：张琼[1] 张莹[1] 白清源[1] 谢丽聪[1] 谢伙生[1] 

机构地区：[1]福州大学数学与计算机科学学院

出　　处：《计算机科学》2008年第3期177-179,共3页Computer Science

基　　金：福州大学科技发展基金资助项目(2005-XQ-13;2006-XQ-22;XRC-0511);福建省教育厅资助项目(JB06023)

摘　　要：传统聚类方法将对象严格地划分到某一类,但很多时候边界对象不能被严格地划分。粗糙集用上近似集和下近似集表示一个类,对这种边界不确定的处理非常有效,典型算法有基于粗糙集的k-means聚类算法和基于粗糙集的leader聚类算法。本文针对RFA(Rough Fuzzy Approach)算法存在的不足,提出了一种新的基于粗糙集的leader聚类算法(NRL,Novel Rough-based Leader)。其基本思想是首先数据项由于与其最近类中心的距离不同,分别被划分到leader集或者supporting leader集,然后对leader集和supporting leader集进行标号,得到聚类结果。实验结果表明NRL算法非常有效。Objects are partitioned into clusters with crisp boundaries in the conventional algorithms. However, clusters do not necessarily have crisp boundaries. Rough set is represented with lower-bound and upper-bound, and is good for the case. At present, there have been some typical algorithms, such as the rough-based k-means clustering algorithm and the rough-based leader clustering algorithm. In this paper, a novel rough-based leader clustering algorithm is pro- posed, since there are some disadvantages in the RFA algorithm. At first, data are partitioned into the set of leaders or the set of supporting leaders according to the difference of the distance of data and the nearest leader. And then, it la-bels the set of leaders and the set of supporting leaders in order to find the clustering result. We present results to demonstrate its validity.

关 键 词：聚类 粗糙集 K-MEANS算法 leader算法 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论] TP18[自动化与计算机技术—计算机科学与技术]