Vehicles Frequently Appearing Together, or VFATs, can be clues in solving criminal cases. Traditional sequence mining approaches help identify VFATs from passing-through records collected at monitoring sites. However, huge traffic data streams hinder fast identification of VFATs. In this paper, we present a multi-threaded approach to fast identification of VFATs based on multi-core processors, called Frequent Sequential Mining based on Multi-Cores (FSMMC). It parallels the execution of tasks, partitions large volumes of data, and obtains VFATs by merging local candidates discovered in different threads running on different processor cores. Through local parallel reduction, FSMMC eliminates the repetitive patterns and reduces computational effort. Moreover, it achieves workload balance by the dynamic distribution of tasks to a pool of threads where the thread that finishes first joins another running thread. Both theoretical analysis and case studies show that FSMMC takes full advantage of multi-core computing platforms and has higher speed-up when searching VFATs among massive passing through records, compared with other approaches without multithreading.
Digital Object Identifier (DOI)
Yu, Dongjin; Dou, Wensheng; Li, Wanqing; Zheng, Suhang; and Shao, Jianhua
"Mining Vehicles Frequently Appearing Together from Massive Passing Records,"
Applied Mathematics & Information Sciences: Vol. 09
, Article 37.
Available at: https://dc.naturalspublishing.com/amis/vol09/iss3/37