基于特征选择的数据流聚类  

Data stream clustering based on feature selection

作　　者：龙鹏飞[1] 唐军[1] 王琳[1] 

机构地区：[1]长沙理工大学计算机与通信工程学院,湖南长沙410114

出　　处：《计算机工程与设计》2010年第19期4235-4237,4241,共4页Computer Engineering and Design

基　　金：国家自然科学基金项目(10871031;60474070);湖南省科技计划基金项目(2008FJ3015)

摘　　要：在数据流聚类时,冗余特征会影响数据的聚类质量,移除冗余特征以提高聚类质量就显得尤为重要。为解决此问题,提出一种基于特征选择的数据流聚类算法(DSCFC)。该算法应用了特征排序、特征等级评定、探测冗余不重要的特征、移除冗余特征算法等。实验结果表明,DSCFC算法能探测出数据流中隐含的冗余特征并移除冗余特征,在对有冗余特征的数据流聚类时,比CluSteam算法更有效,聚类质量更好。Clustering in the data stream, the redundant features will affect the quality of data clustering, removing redundant features to improve the clustering quality is very important, To solve this problem, it is proposed that a data stream clustering algorithm based on feature selection （DSCFC）. It is one-pass clustering algorithms, these are applied that ranking feature, grading feature, detecting redundant features and removing the redundant features algorithm and so on. The experimental results indicated that DSCFC algorithm can detect hidden redundant features in data stream and remove redundant features; when there are redundant features in the data stream clustering, the algorithm is more efficient than CluStream, clustering quality is better.

关 键 词：数据流聚类 特征选择 冗余特征 代价矩阵 特征移除 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]