This application presents an improved approach to predict the public arrival time based on location. After analyzing the components of arrival time systematically, the arrival time and dwell time at previous stops are chosen as the main input variables of the prediction model.