DBSCAN: Past, present and future

Data Mining is all about data analysis techniques. It is useful for extracting hidden and interesting patterns from large datasets. Clustering techniques are important when it comes to extracting knowledge from large amount of spatial data collected from various applications including GIS, satellite images, X-ray crystallography, remote sensing and environmental assessment and planning etc. To extract useful pattern from these complex data sources several popular spatial data clustering techniques have been proposed. DBSCAN (Density Based Spatial Clustering of Applications with Noise) is a pioneer density based algorithm. It can discover clusters of any arbitrary shape and size in databases containing even noise and outliers. DBSCAN however are known to have a number of problems such as: (a) it requires user´s input to specify parameter values for executing the algorithm; (b) it is prone to dilemma in deciding meaningful clusters from datasets with varying densities; (c) and it incurs certain computational complexity. Many researchers attempted to enhance the basic DBSCAN algorithm, in order to overcome these drawbacks, such as VDBSCAN, FDBSCAN, DD_DBSCAN, and IDBSCAN. In this study, we survey over different variations of DBSCAN algorithms that were proposed so far. These variations are critically evaluated and their limitations are also listed.