Critical analysis of DBSCAN variations
Tariq M. Ali, Sohail Asghar, Naseer Ahmed Sajid · 2010
DBSCAN is a widely used technique for clustering in spatial databases. DBSCAN needs less knowledge of input parameters. Major advantage of DBSCAN is to identify arbitrary shape objects and removal of noise during the clustering process. Beside its familiarity, DBSCAN has problems with handling large databases and in worst case its complexity reaches to O(n2). Similarly, DBSCAN cannot produce correct result on varied densities. Some variations are proposed to DBSCAN, to show its working in some other domains. In this paper we surveyed some important techniques in which original DBSCAN is modified or enhanced with improvement in complexity or result improvement on varied densities. We define criteria and analyse these variations with complexity (time and space) and output to the original DBSCAN algorithm. We also compare these variations with one another to select the efficient algorithm. In most of the variations partitioning and hybrid methodologies are originated to deal DBSCAN problems. We concluded with some variations which perform better than other variation over defined criteria (objectives).