Cluster Ensemble Using Feature Selection and Sample Subspace
In cluster analysis area, cluster ensemble method performs better than traditional clustering algorithms. Diversity plays an important role in cluster ensemble, this paper introduces a variant of generic cluster ensemble algorithm where the number of clusters in ensemble member is specified in an interval. The ensemble can be used for feature-distributed clustering, each clustering partition has access to only a limited number of features of each pattern, and different data subsets can form different clusters, thus combing these multiple clustering partitions can provide a more stable cluster structure. So, go further, we use feature selection and sample subspace in cluster ensemble, which can keep diversity and reduce the ensemble complexity. Experiments on both real data sets and synthetic data sets show the advantage of using diversity in cluster ensemble.
Cluster ensemble Feature selection resampling
Guiling Li Jinlei Li Xiaolian Zhu
College of Computer Science and Technology, Huazhong University of Science and Technology, Wuhan, 43 School of Computer Science, China University of Geosciences, Wuhan, 430074,China
国际会议
武汉
英文
2007-09-21（万方平台首次上网日期，不代表论文的发表时间）