GANNSTER: Graph-augmented neural network spatio-temporal reasoner for traffic forecasting

Traffic forecast is a problem of high interest due to its impact on mobility and inherent socio-economic aspects of people’s lives. Particularly for adaptive traffic light systems, the ability to predict traffic throughput in intersections enables fast adaptation, thus reducing traffic jams. In this work, we propose a novel approach for traffic forecasting, termed Graph Augmented Neural Network Spatio-TEmporal Reasoner (GANNSTER), which fuses spatial information, given by the traffic network topology, with temporal reasoning and learning capabilities of recurrent neural networks. Our modelling contribution is supplemented by the public release of a novel real-world dataset containing urban traffic throughput in intersections. We comparatively evaluate GANNSTER against state-of-the-art models for traffic forecast and demonstrate its superior performance.