Periodic Pattern Detection Algorithms for Personal Trajectory Data Based on Spatiotemporal Multi-Granularity

Identifying periodic patterns in individuals' trajectories is the basis of location awareness and personalized location services. It can help us understand personal behaviors. However, fuzziness and uncertainty of trajectory data, as well as noise and period distortion, make it difficult to recognize periodic patterns. In addition, the period lengths are usually unknown, and patterns have multiple granularities. Most of the existing mining algorithms focus on the discovery of patterns with different period lengths at a specific spatial scale, and few algorithms identify periodic patterns from the perspective of spatiotemporal multi-granularity. Based on the existing studies, we propose a framework for identifying periodic patterns with different spatiotemporal granularities from personal trajectory data. First, a sequence of trajectory points is transformed into a time series of locations by using trajectory abstraction. Then, a multi-granularity behavior model is defined from spatial and temporal information. Finally, the single behavior periodic patterns can be discovered without knowing the period length by using a novel algorithm. Based on the association rules between locations, we can determine the periodic patterns of multiple behaviors from single behavior patterns. To evaluate the accuracy and efficiency of the algorithms, an artificial trajectory sequence and a real-world trajectory dataset called GeoLife are used in comparative experiments. The experimental results show that the proposed algorithm has higher accuracy on the promise of efficiency.