摘要
软子空间聚类是高维数据分析的一种重要手段.现有算法通常需要用户事先设置一些全局的关键参数,且没有考虑子空间的优化.提出了一个新的软子空间聚类优化目标函数,在最小化子空间簇类的簇内紧凑度的同时,最大化每个簇类所在的投影子空间.通过推导得到一种新的局部特征加权方式,以此为基础提出一种自适应的k-means型软子空间聚类算法.该算法在聚类过程中根据数据集及其划分的信息,动态地计算最优的算法参数.在实际应用和合成数据集上的实验结果表明,该算法大幅度提高了聚类精度和聚类结果的稳定性.
Soft subspace clustering is a key for high-dimensional data analysis. The existing algorithms usually require users to estimate some key global parameters in advance, and ignore the optimization of subspaces. A novel objective function, to be optimized by the soft subspace clustering algorithms, is proposed in this paper by taking into account both minimization of the compact subspace clusters and maximization of the subspaces in which the clusters exist. Based on this, a new locally feature weighting scheme is derived, and an adaptive algorithm for k-means type soft subspace clustering is presented. In the new algorithm, the optimal values of parameter are automatically computed, according with the dataset and its partitions. Experimental results carried out on some real-world and synthesis datasets demonstrate that the proposed method significantly improves the accuracy as well as the stability of the clustering results.
出处
《软件学报》
EI
CSCD
北大核心
2010年第10期2513-2523,共11页
Journal of Software
基金
国家自然科学基金No.10771176
福建省自然科学基金No.2009J01273
国家教育部留学回国人员科研启动基金No.[2008]890
福建省省属高校科研专项重点项目No.JK2009006
关键词
聚类
高维数据
子空间
特征加权
自适应性
clustering
high dimensional data
subspace
feature weighting
adaptability
