AGED (Automatic Generation of Eps for DBSCAN) Neha Soni 1 , Dr. Amit Ganatra 2 1 Computer Engineering Dept. SVIT, Vasad, Gujarat, India 2 Faculty of Tech. & Engg., CHARUSAT, Changa, Gujarat, India Abstract DBSCAN is one of the widely used density based algorithm for clustering in data mining. However, the two major challenges in DBSCAN are, firstly, the tuning of input parameter Eps (Neighborhood radius) and secondly, the handling of the varied density datasets. A minor change in value of Eps will affect the clustering, moderately or highly with respect to dataset distribution. For varied density datasets, the global value of Eps will not be able to generate proper clusters. This paper proposes a new algorithm AGED for automatic generation of multiple Eps. It generates a set of values of Eps for different density levels that may exist in the dataset. Then, with different values of Eps, depending on the type of dataset either varied or non-varied, DBSCAN algorithm is adopted for finding the clusters. The automatic generation of Eps leads to better clustering with lesser efforts in tuning of the same. The experimental results are also promising for the two dimensional datasets and multidimensional non-varied density datasets as well as varied density datasets. Keywords: DBSCAN, varied density datasets, parameter selection, Eps, density based clustering 1. Introduction Cluster Analysis is a process of grouping the objects where the objects can be physical for e.g. a chair or can be abstract objects such as behavior of a customer, handwriting. Clusters are formed based on the similarity or dissimilarity among the objects given in specific region. Similar objects belong to the same cluster. Many algorithms have been proposed for clustering, each with the inclination to address a specific issue/s like dimensionality, scaling, handling outliers etc. However, there is not a single generalized algorithm to address all the issues together [4]. This paper proposes an algorithm that addresses both the major challenges of DBSCAN stated above. Different algorithms proposed may follow good features of the different methodology and thus it is difficult to categorize them with a solid boundary. The detailed categorization of the clustering algorithms is given in [1-4, 28]. All clustering algorithms, basically, can be categorized into two broad categories: partitioning and hierarchical, based on the properties of generated clusters [2-3]. International Journal of Computer Science and Information Security (IJCSIS), Vol. 14, No. 5, May 2016 536 https://sites.google.com/site/ijcsis/ ISSN 1947-5500