Autonomous driving vehicles must have the ability to identify and predict behaviors of surrounding moving objects (e.g., other vehicles, cyclists, and pedestrians) in real-time. This is especially true in urban environments, where interactions become more complex due to high volumes of traffic. The work in this paper harnesses the Dempster-Shafer (DS) theoretic framework's ability to capture and account for various types of evidence uncertainty to develop a robust event detection and prediction model, which is appropriately calibrated to account for the underlying uncertainty so that it may be employed to arrive at a more informed decision.