Title: G-OPTICS: fast ordering density-based cluster objects using graphics processing units

Authors: Wookey Lee; Woong-Kee Loh

Addresses: Department of Industrial Engineering, Inha University, Incheon 22212, South Korea ' Department of Software, Gachon University, Seongnam 13120, South Korea

Abstract: Clustering is the process of forming groups or clusters of similar objects in the dataset and has been used as an important tool for many data mining applications including the web-based ones. While density-based clustering algorithms are widely adopted, their clustering result is highly sensitive to parameter values. The OPTICS algorithm presents a solution to this problem; it produces an ordering of objects that is equivalent to the clustering results for a wide range of thresholds ϵ. In this paper, we propose an algorithm named G-OPTICS to significantly improve the performance of OPTICS using a graphics processing unit (GPU). The experimental results using real and synthetic datasets demonstrated that G-OPTICS outperformed the previously fastest FOPTICS algorithm by up to 118.3 times (67.7 times on the average).

Keywords: ordering of objects; density-based clustering; GPU; graphics processing unit; parallel algorithm.

DOI: 10.1504/IJWGS.2018.092583

International Journal of Web and Grid Services, 2018 Vol.14 No.3, pp.273 - 287

Available online: 22 Jun 2018 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article