Attention-based recurrent neural network for traffic flow prediction

Traffic flow prediction is an important while complex problem in transportation modeling and management. Many uncertain, non-linear and stochastic factors could have large influence on the prediction performance. With the recent development in deep learning, researchers have applied deep neural networks for the traffic flow prediction problem and achieved promising results. However, existing studies still have some issues unaddressed, e.g., the models only predict the traffic flow at next time step while travelers may need a sequence of predictions to make better, long-term decisions; temporal factors are (e.g., day of the week, national holiday) usually not well considered during prediction. To address these limitations, this paper proposed an attention-based recurrent neural network architecture for multi-step traffic flow prediction. Experimental results demonstrate that the proposed method has superior performance compared to the existing models. We also show how the method can be used to develop traffic anomaly detection systems.