Neural Congestion Prediction System For Trip Modelling In Heterogeneous Spatio-Temporal Patterns

Until recently, urban cities have faced an increasing demand for an efficient system able to help drivers to discover the congested roads and avoid the long queues. In this paper, an Intelligent Traffic Congestion Prediction System (ITCPS) was developed to predict traffic congestion states in roads. The system embeds a Neural Network architecture able to handle the variation of traffic changes. It takes into account various traffic patterns in urban regions as well as highways during workdays and free-days. The developed system provides drivers with the fastest path and the estimated travel time to reach their destination. The performance of the developed system was tested using a big and real-world Global Positioning System (GPS) database gathered from vehicles circulating in Sfax city urban areas, Tunisia as well as the highways linking Sfax and other Tunisian cities. The results of congestion and travel time prediction provided by our system show promise when compared to other non-parametric techniques. Moreover, our model performs well even in cross-regions whose data were not used during training phase.