TrailMarker: Automatic Mining of Geographical Complex Sequences

Given a huge collection of vehicle sensor data consisting of d sensors for w trajectories of duration n, which are accompanied by geographical information, how can we find patterns, rules and outliers? How can we efficiently and effectively find typical patterns and points of variation? In this paper we present TRAILMARKER, a fully automatic mining algorithm for geographical complex sequences. Our method has the following properties: (a) effective: it finds important patterns and outliers in real datasets; (b) scalable: it is linear with respect to the data size; (c) parameter-free: it is fully automatic, and requires no prior training, and no parameter tuning. Extensive experiments on real data demonstrate that TRAILMARKER finds interesting and unexpected patterns and groups accurately. In fact, TRAILMARKER consistently outperforms the best state-of-the-art methods in terms of both accuracy and execution speed.