An Efficient and Fast Density Conscious Subspace Clustering using Affinity Propagation

Subspace clustering is an eminent task to detect the clusters in subspaces. Density-based approaches assume the high-density region in the subspace as a cluster, but it creates density divergence problem. The proposed work improves the performance of Density Conscious subspace clustering (DENCOS) by utilizing the Affinity Propagation (AP) algorithm to detect the local densities for a dataset. Initially, the Affinity Propagation based Sampling Algorithm was used to coarsen the input sparse dataset and choose a small number of f inal representative exemplars. The DENCOS is used to partition those exemplars and finally the cluster assignment of all data points is achieved through their corresponding representative exemplar. On experimentation, it is found that the proposed algorithm outperforms the original DENCOS algorithm and other subspacing algorithms in terms of speed, memory usage and quality of subspace clustering. Index Terms —Data Mining, Data Clustering, Density Conscious Subspace Clustering, Affinity Propagation.