Probabilistic Graphical Model based Personal Route Prediction in Mobile Environment

Individuals tend to follow their own preferred paths when traveling to specific places. Information on these routes could be utilized to build various intelligent LBSs . In order to predict a current user’s route, various approaches have been researched. In this paper, we suggest a practical approach to learning users' route patterns from their histories and using that information to predict specific routes. In cases where existing routes overlap, i.e., where parts of routes are the same, in a user's route model, it is difficult to identify the user's intended path. For more accurate prediction, firstly, we extract route patterns by adopting image processing. Secondly, we build a stateobservation model reflecting users' intentions, based on route patterns, temporal features and weather information. Our approach consist of four steps: recognizing regions for splitting routes into trip segments, route pattern mining, learning users' route models and trip route prediction. Our method achieved a prediction accuracy of 96.4% in tests performed with 15 smartphone users.