Automatic Mining of Multi-granularity Temporal Regularity from Trajectory Data.

Siyuan Huang,Rui Zhang,Nuofei Li,Jiming Guo,Hongbo Jiang

DOI: https://doi.org/10.1145/3148055.3148067

2017-01-01

Abstract:Temporal regularity in trajectory data is an important basis for traffic management, public service and marketing. Although many efforts have been made to study temporal regularity, yet almost all existing works select time granularity intuitively. User-specified time granularity and other parameters may lead to biased results. Moreover, as the size of datasets grows, the costs of parameters tuning also increases. To solve these problems, we propose the Automatic Multi-granularity Temporal Regularity Detection algorithm (auto-MTRD) for trajectory data. Our approach clusters time series from the trajectory data using automatic parameter selection and generates a temporal regularity tree to indicate multi-granularity temporal regularity. It cannot only avoid the negative effect of human intervention, but also evaluate the relative importance of multiple time granularities at the same time. Two real-life datasets are used to validate the effectiveness of our method.