Predicting Smartphone Users’ Future Locations through Deep Reinforcement Learning

This paper demonstrates how people's future movement patterns can be predicted by learning from their location history from past several weeks. We introduce a novel scheme of applying deep reinforcement learning model on a relatively recent, real-world location history dataset collected from ordinary users' smartphone daily usage. Through a collaborative project between our affiliated university and commercial companies which distribute smartphone applications, 7,236 users of those applications agreed to continuously share their location history and participated over a period of 3 months in year 2017. Our experiments on this dataset reveals that deep reinforcement learning can be successfully applied to predict locations where users are likely to visit in the near future.