A new generic interpretation of enhanced subspace clustering in high dimensional data
Jyotsna Devi, M. Venkateshwara Rao · 2018
The prominent challenging task in data mining is to find the high dimensional data point clusters. In particular, the subspace clustering methods can be understood well in high dimensional data mining process. However the traditional subspace clustering techniques failed to find significance and quality of clusters that are present in the identified subspaces in growing the number of dimensions in large data. Most of the conventional clustering algorithms used bottom up search method and took multiple database scans to lead inefficiency. This research paper focuses a new enhanced subspace clustering scheme called ENSUBCLU, which overcomes the inefficiency from traditional subspace clustering techniques. Initially ENSUBCLU model was found in the dense units for each one dimensional projection of a given dataset. After that, applied subspace steering scheme to identify the promising subspaces and their combinations of common points among one dimensional subspaces. This model finds all interesting combinational dense core regions, from all lower dimensions of dense units. This lead to the reduction of subspace processing and obtain high quality subspace clusters and eliminates the redundant subspace clusters using hashing technique. Finally this model scales well with increasing attributes. ENSUBCLU model presents an empirical study on various synthetic, real world datasets and find the maximal subspace clusters in more improved manner than existing algorithms. It can even tackle many application areas like social networking, computer vision, bio-informatics, financial and sales analysis maintaining the high dimensional data.