Abstract:

We propose an efficient distributed mining algorithm based on MapReduce for mining swarm pattern from big spatiotemporal trajectory data. We first define the objectclosed swarm pattern based on the maximum moving object set, and optimize the serial mining algorithm using the strategy of minimum time support set to minimize the computation costs. We then propose a parallel swarm mining model based on the time independence, and the clustering and the objectclosed swarm mining on the time domain are parallelized. Finally, we propose a distributed mining algorithm based on MapReduce chained architecture, which quickly discovers swarm patterns in big trajectory data by a 4stage framework. Experimental evaluations on the Hadoop platform, using massivescale real world traffic trajectory datasets, demonstrate the effectiveness and efficiency of the proposed distributed algorithm.

Key words: spatiotemporal trajectory mining;big data;swarm pattern;distributed;MapReduce