Subspace Clustering Based On Dimension-oriented Distance And Its Applications

In high-dimensional data clustering analysis,due to the increasing of dimension,the traditional method is difficult to be applied effectively.To solve this problem,the traditional approach is to delete some unimportant variables,or variables linear combination so as to achieve the purpose of dimension reduction.Sometimes since the cluster is formed in different subspace,traditional dimension reduction methods may delete important variable,which leads to the loss of useful information.Therefore,it is very useful to identify the subspace of each class for high-dimensional data.The subspace clustering(DSC)method based on dimension distance is exploited in this paper,it can be used to dimension reduction without losing information,as well as automatically identify the number of clusters.The method has two core ideas:(1)Dimension-oriented distance(dod),which utilizes the information of the number of variables and the information of variables as an intrinsic element of distance.(2)Dimension reduction based on gap statistic.It can identify automatically and effectively the subspace of each class.Some real data are explored to illustrate the implement the method.The comparisons with the traditional method and other subspace clustering methods,show the superiority of our proposed method in high-dimensional data clustering.