PDBPSO: A parallel density clustering algorithm based on ball-tree and IPSO algorithm

In order to deal with the problems of unreasonable data partition, easily trapped into local optima and low parallel performance in density clustering algorithm based on big data, this paper propose a parallel density clustering algorithm based on ball-tree and IPSO algorithm, noted as PDBPSO. This algorithm consists of three stages: data division, local clustering, and global clustering. Firstly, based on ball-tree, we propose an adaptive division strategy (BDG) to divide the dataset adaptively. Secondly, we propose an improve particle swarm optimization algorithm (IPSO) which use the adaptively step strategy based on knowledge learn (AS) and the particle decision function (PF). Then, we use IPSO algorithm to select the optimal parameters of local clustering dynamically, which can improve the clustering effect of local clustering. Meanwhile, in order to improve the parallel efficiency, we propose a density-based clustering algorithm using IPSO algorithm (DBIPSO) to parallel compute the local clusters of clustering algorithm. Finally, based on MapReduce and disjoint set, we propose the parallel local clusters merging algorithm (MECR) to compute the result of clustering algorithm faster. The experimental results show that the PDBPSO algorithm has better clustering results and performs better parallelization in big data.