Subspace clustering for high dimensional datasets

Clustering high dimensional data is a challenging problem because of the existence of many irrelevant and redundant attributes. Conventional clustering algorithms identify a global set of relevant attributes prior to clustering using attribute selection and feature extraction techniques. All the globally relevant attributes are used in the similarity calculation while clustering. These algorithms fail to identify true clusters that are present in a subset of attributes. So, subspace clustering has become the thrust area of research in the recent past. Subspace clustering detects the clusters that exist in subsets of dimensions. Different types of subspace clustering algorithms are proposed in the literature. This paper discusses the different types of subspace clustering algorithms with main emphasis on 2D subspace clustering. Availability of new and huge datasets like spatiotemporal datasets, temporal datasets, spatial datasets and genomic data has necessitated the development of 3D subspace clustering. This paper presents an overview of subspace clustering for the research community who is interested in subspace clustering.