Exploiting GPU parallelism in improving bees swarm optimization for mining big transactional databases

This paper investigates the use of GPU (Graphics Processing Unit) in improving the bees swarm optimization metaheuristic performance for solving the association rule mining problem. Although this metaheuristic proved its effectiveness, it requires huge computational resource when considering big databases for mining. To overcome this limitation, we develop in this paper a GPU-based Bees Swarm Optimization Miner (GBSO-Miner) where the GPU is used as a co-processor to compute the CPU-time intensive steps of the algorithm. Unlike state-of-the-art GPU-based ARM methods, all BSO steps including the determination of search area, the local search, the evaluation, and the dancing are performed on GPU. A mapping method between the data input of each task and the GPU blocks/threads is developed. To demonstrate the effectiveness of the GBSO-Miner framework, intensive experiments have been carried out. The results show that GBSO-Miner outperforms the baseline methods of the literature (GPApriroi, MEGPU, and Dmine) using big textual and graph databases. The results reveal that GBSO-Miner is up to 800 times faster than an optimized CPU-Implementation.