Traffic flow prediction with Long Short-Term Memory Networks (LSTMs)

Accurate traffic flow information is crucial for an intelligent transportation system management and deployment. Over the past few years, many existing models have been designed for short-term traffic flow prediction. However, they fail to provide favorable results due to their shallow architectures or incapability to extract the sequence correlations in the data. In this paper, we explore the application of Long Short-Term Memory Networks (LSTMs) in short-term traffic flow prediction. As a deep learning approach, LSTMs are able to learn more abstract representations in the non-linear traffic flow data. The intrinsic feature of capturing long-term dependencies in a sequential data also makes it a suitable choice in traffic prediction. Experiments on real traffic data sets indicate a good performance of our model. The LSTMs architecture is also compared with state-of-the-art models and experiments show that our model achieves desirable results by lowering the MAPE metrics to 5.4%.