Density-based clustering algorithm for GPGPU computing
Kai-Shiang Chang, Yi-Wen Peng, Wei‐Mei Chen · 2017
Clustering is a common data mining procedure that groups multi-dimensional points with similar components to form different subsets. Among all of the clustering algorithms, DBSCAN is one of the most popular algorithms owing to finding clusters with arbitrary shapes and noise of datasets. However, with data volumes growing and the execution time of algorithms becoming longer, numerous methods have been developed to accelerate algorithm execution times of DBSCAN. This paper proposed a GPU-based DBSCAN algorithm that exploits the points with similar components to establish indices, thereby facilitating the identification of core points fast. In addition, this algorithm specified search criteria for indices search and clustered data through the parallel partial connection to the core point to reduce the amount of distance computation between the points. The experimental results indicated that the proposed algorithm was more efficient than the other GPU-based DBSCAN.