ECR-DBSCAN: An improved DBSCAN based on computational geometry

A new density based clustering algorithm ECR−DBSCAN based on DBSCAN, has been presented in this paper. Computational geometry is applied to develop the modified DBSCAN algorithm. It is well known that the quality of density based clustering depends on its input parameters. However, it is not easy to determine proper values of input parameters for DBSCAN. This paper presents three significant modifications or extensions to DBSCAN related with (i) selection of hyper parameter epsilon (eps) using the radii of empty or voronoi circles (ii) selection of parameter minPoints (mp) for the same epsilon and (iii) redistribution of noise points to suitable clusters using the concept of centroid hinged clustering. ECR−DBSCAN is implemented with PYTHON accompanied by extensive experiment on benchmark data sets. Our experimental results establish the novelty and validity of the proposed clustering method over standard techniques.