Traffic congestion prediction and detection employ a vital role in intelligent transportation networks. The main goal of real-time traffic prediction is to conduct data processing and traffic condition assessment quickly. Traffic congestion prediction provides early traffic jam detection, route planning, guide vehicle transmission, and reduce congestion. Traffic data is Spatio-temporal, and it is constantly changing with time and space. Moreover, it is complex and has dynamic Spatio-temporal dependencies. Traffic congestion predictions are performed based on the recurring congestion and non-recurring congestion. Deep learning models exploit more features and complex architectures than the classical methods of traffic prediction and achieve better performance.
Modeling Spatial Dependency and Modeling Temporal Dependency are the two methods of deep learning involved in traffic congestion prediction. Flow, demand, speed, travel time, and occupancy are the application tasks involved in traffic congestion prediction. Convolutional neural networks and long short-term memory are the commonly used deep learning architectures in traffic congestion prediction. Future research directions of traffic congestion prediction on deep learning are synergies between model-driven and deep learning approaches, traffic prediction with high dimensionality, long-term traffic congestion prediction, integration of transfer learning to deep Spatio-temporal models for a few shot problems, real-time traffic prediction with lightweight neural network and many more.