Gilad Katz

While location prediction of smartphone users has made great strides in recent years, a major challenge remains. As users spend the majority of their time is several fixed locations (home, work), existing algorithms are unable to identify the exact time in which a person is likely to depart from one place to another. In this work we present a sensor-based approach designed to predict the departure time of users. By using location and accelerometer sensors we were able to train a generic classification model that is able to predict whether the user will stay put or move to a different location with true positive rate of 0.73 and false positive rate of 0.3.