Location prediction: a deep spatiotemporal learning from external sensors data

This paper proposes a deep multi-task learning framework to predict the next location from trajectories that are captured by external sensors (e.g., traffic surveillance cameras, or speed radars). The reported positions in such trajectories are sparse, due to the sparsity of the sensor distribution, and incomplete, because the sensors may fail to register the passage of objects. In this framework, we propose different preprocessing steps to align the trajectories representation and cope with a missing data problem. The multi-task learning approach is based on Recurrent Neural Networks. It utilizes both time and space information in the training phase to learn more meaningful representations, which boosts the learning performance of location prediction. The multi-task learning model, together with the preprocessing step, substantially improves the prediction performance. We conduct several experiments using a real dataset, and they demonstrate the validity of our multi-task learning model in terms of accuracy of 85.20%, which is more than 20% better than using a single-task learning model.