Evaluating Deep and Classical Learning Models for Spatiotemporal GPS Trajectory Forecasting with Reliable Uncertainty Bounds