Modified DBSCAN Clustering Algorithm for Data with Different Densities.

Abstract

The problem of detecting clusters of points in data is challenging when the clusters are of different size, density and shape. The density based clustering algorithm DBSCAN is one of the most popular density based algorithms. The DBSCAN algorithm has a limitation when dealing with data of different densities. In this paper we propose an algorithm based on the DBSCAN. The proposed algorithm is capable of clustering data with arbitrary shapes and dealing with different densities of data. The Idea of the proposed algorithm is to update the eps and MinPts (where eps and MinPts are input parameters of DBSCAN algorithm) values according to the densities of regions of data points. These values are scaled depending on eps-neighborhood points. In the experiments we apply the proposed algorithm to artificial dataset and real dataset as we will show in the last section of the paper.