摘要
在处理高维数据时,聚类的工作往往归结为对子空间的划分问题。大量的真实实验数据表明,相同的属性对于高维数据的每一类子空间而言并不是同等重要的,因此,在FCM算法的基础上引入了方差权重矩阵模型,创造出了新的聚类算法称之为WM-FCM。该算法通过不断地聚类迭代调整权重值,使得其重要的属性在各个子空间内更为显著地表征出来,从而达到更好的聚类效果。从基于模拟数据集以及UCI数据集的实验结果表明,该改进的算法是有效的。
In dealing with high-dimensional data, clustering can be viewed as finding out an appropriate subspace division. However, lots of real experimental data show that for different classes of the high dimensional data subspaces, the same attributes are not equally important. This paper presented the new high dimensional subspace clustering algorithm WM-FCM, which integrated the FCM clustering algorithm with the proposed variance weight matrix model. Through continuous clustering iterations,the algorithm adjusted the weights of attributes of each subspaee so that important attributes became more significant, thus led to better performance of subspace clustering. The experimental results on artificial data sets and UCI （ University of California,Irvine） data sets show that the presented algorithm WM-FCM is effective.
出处
《计算机应用研究》
CSCD
北大核心
2012年第8期2868-2871,2881,共5页
Application Research of Computers
基金
国家自然科学基金资助项目(90820002)
江苏省自然科学基金资助项目(BK2009067)
