MRG-DBSCAN: An Improved DBSCAN Clustering Method Based on Map Reduce and Grid

Abstract

DBSCAN is a density-based clustering algorithm. This algorithm clusters data of high density. The traditional DBSCAN clustering algorithm in finding the core object, will use this object as the center core, extends outwards continuously. At this point, the core objects growing, unprocessed objects are retained in memory, which will occupy a lot of memory and I/O overhead, algorithm efficiency is not high. In order to ensure the high efficiency of DBSCAN clustering algorithm, and reduce its memory footprint. In this paper, the original DBSCAN algorithm was improved, and the G-DBSCAN algorithm is proposed. G-DBSCAN algorithm reduces the number of query object as a starting point. Put the data into the grid, with the center point of the data in the grid to replace all the grid points as the algorithm input. The query object will be drastically reduced, thus improving the efficiency of the algorithm, reduces the memory footprint. In order to make the G-DBSCAN algorithm can adapt to large data processing, we will parallelize the G-DBSCAN algorithm, and combining it with Map Reduce framework. The results prove that G-DBSCAN and MRG-DBSCAN algorithm are feasible and effective.