Abstract
Prediction of traffic conditions plays a key role in the current era's intelligent transportation system. It not only enables commuters to choose appropriate routes to reach their destinations but also helps authorities in making effective traffic management plans. All needed for this purpose is to use a method that could handle abundant traffic data and by making wise use of this data, it could help authorities to get an estimate of traffic conditions on road networks and to make effective traffic management plans. Deep learning approaches are the most appropriate choice for these kinds of problems and have extensively been used for traffic forecast. In this work, we are using deep neural networks to predict the traffic condition on highways by considering the spatio-temporal correlation in traffic data attributes. For our deep model, we are using historical traffic data collected from Performance Measurement System (PeMS) for a period of three months on a selected patch of a highway in California. We are using vehicles occupancy values collected from the vehicle detector stations (VDSs) to predict the occupancy on the freeway. Prediction results compared with the actual occupancy values not only give very high accuracy but also enables us to make use of these predicted values in performing different traffic management tasks.