Grid-Based Approach to Determining Parameters of the DBSCAN Algorithm.

Clustering is a very important technique used in many fields in order to deal with large datasets. In clustering algorithms, one of the most popular approaches is based on an analysis of clusters density. Density-based algorithms include different methods but the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) is one of the most cited in the scientific literature. This algorithm can identify clusters of arbitrary shapes and sizes that occur in a dataset. Thus, the DBSCAN is very widely applied in various applications and has many modifications. However, there is a key issue of the right choice of its two input parameters, i.e the neighborhood radius (eps) and the MinPts. In this paper, a new method for determining the neighborhood radius (eps) and the MinPts is proposed. This method is based on finding a proper grid of cells for a dataset. Next, the grid is used to calculate the right values of these two parameters. Experimental results have been obtained for several different datasets and they confirm a very good performance of the newly proposed method.