Trajectory pattern mining with multistage spatial partitioning

Abstract

Most trajectory pattern mining techniques assume that the data to be analyzed contain complete and evenly distributed spatial and temporal information. However in reality, collected data may contain noise, missing or incomplete information, and uneven spatial resolution. In trajectory pattern mining methods, trajectory patterns are extracted by splitting spatial workspace into uniformly tiny sized squares, followed by determining popular cells which contain many data points. Finally, a sequential pattern mining technique, e.g. MiSTA, is used to extract trajectory pattern. This research proposes non-uniform partitioning to handle uneven spatial distribution as modification towards the uniform spatial workspace division process. The proposed approach, named multistage spatial partitioning is developed based on point-region quadtree concept. The new partitioning method is conducted for preprocessing before applying MiSTA. As the result, using multistage spatial partition, MiSTA succeeds in uncovering more detailed and broader coverage patterns compared to uniform partitioning approach through a series of experiments.