一种基于滑动窗口的不确定数据流聚类算法  

Clustering algorithm over uncertain data streams based on sliding window

作　　者：朱娟芳[1] 霍欢[1] 徐亚[1] 奚金金[1] 彭敦陆[1] 高丽萍[1] 

机构地区：[1]上海理工大学光电信息与计算机学院,上海200093

出　　处：《信息技术》2013年第4期1-5,共5页Information Technology

基　　金：国家自然科学基金(61003031);上海市研究生创新基金项目(JWCXSL1202)

摘　　要：针对不确定数据流上的聚类问题提出了一种新的聚类算法SWCUS,它采用滑动窗口缓存一段时间内的元组作为聚类对象,只存储新到达的数据,"淘汰"过时数据,以便获得高质量的聚类结果。另外它还应用k-means算法来生成初始微簇并且提出了一种新的离群点机制来排除离群点。实验结果表明,SWCUS算法与同类型的算法相比有较好的聚类效果和较快的聚类速度,而且其自身拥有很强的可伸缩性。This paper proposes a new clustering algorithm SWCUS for solving clustering problem over the uncertain data stream, which uses sliding window to save tuples in a period of time as clustering objects. In order to obtain high quality clustering results, it only store new arrived data, ＂elimination＂ out of date data. In addition, it uses k-means algorithm to generate initial clusters and presents a new oufliers mechanism to remove outliers. The experiments show that the SWCUS algorithm has better clustering effect and faster clustering speed than same type algorithms. It also has greater expandability.

关 键 词：不确定数据流 聚类 滑动窗口 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]