A New Approach for Subspace Clustering of High Dimensional Data

Abstract

Clustering high dimensional data is an emerging research area. The similarity criterion used by the traditional clustering algorithms is inadequate in high dimensional space. Also some of the dimensions are likely to be irrelevant thus hiding a possible clustering. Subspace clustering is an extension of traditional clustering that attempts to find clusters in different subspaces within a dataset. This paper proposes an idea by giving weight to every node of a cluster in a subspace. The cluster with greatest weight value will have more number of nodes when compared to all other clusters. This method of assigning weight can be done in two ways such as top down and bottom up. A threshold value is fixed and clusters with value greater than threshold only will be taken into consideration. The discovery of clusters in selected subspaces will be made easily with the process of assigning weight to nodes. This method will surely result in reduction of search space.