Calculation of Density-Based Clustering Parameters Supported with Distributed Processing

Abstract

In today’s world of data-mining applications there is a strong need for processing spatial data. Spatial objects clustering is often a crucial operation in applications such as traffic-tracking systems or telemetry-oriented systems. Our current research is focused on providing an efficient caching structure for a telemetric data warehouse. We perform spatial objects clustering for every level of the structure. For this purpose we employ a density-based clustering algorithm. However efficient and scalable, the algorithm requires an user-defined parameter Eps. As we cannot get the Eps from user for every level of the structure we propose a heuristic approach for calculating the Eps parameter. Automatic Eps Calculation (AEC) algorithm analyzes pairs of points defining two quantities: distance between the points and density of the stripe between the points. In this paper we describe in detail the algorithm operation and interpretation of the results. The AEC algorithm was implemented in both centralized and distributed version. Included test results compare the two versions and verify the AEC algorithm correctness against various datasets.