A Frequent Pattern based Prediction Model for Moving Objects

Abstraction Huge amounts of moving object data have been collected with the advances in wireless communication and positioning technologies. Trajectory patterns extracted from historical trajectories of moving objects can reveal important knowledge about movement behavior for high quality LBS services, especially for location prediction. Existing approaches cannot forecast accurate locations in the distant future since they use motion functions which emphasize the recent movements of objects. In this paper, we propose a new approach which utilizes frequent trajectory patterns to predict location. Using line simplification and clustering, the proposed method simplifies trajectories and clusters them into spatio-temporally meaningful regions. After original trajectories are discretized into the sequences using regions, trajectory patterns from discretized sequences are extracted using a prefix-based projection approach. Then, we construct a tree-structured prediction model using these patterns, which allows an efficient indexing of discovered patterns to find the best match. We experimentally analyze that the proposed method’s efficiency in discovering trajectory patterns, predicting a future location accurately even though the query time is far in the future.