Hybrid graph convolution neural network and branch-and-bound optimization for traffic flow forecasting

• Published in: Future generation computer systems Vol. 139; pp. 100 - 108
• Main Authors: Djenouri, Youcef, Belhadi, Asma, Srivastava, Gautam, Lin, Jerry Chun-Wei
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.02.2023
• Subjects: Deep learning; Forecasting; Graph convolution neural network; Smart roads; Deep learning; Smart roads; Graph convolution neural network; Forecasting
• ISSN: 0167-739X; 1872-7115
• DOI: 10.1016/j.future.2022.09.018

In this study, we combine graph optimization and prediction in a single pipeline to investigate an innovative convolutional graph-based neural network for urban traffic flow prediction in an edge IoT environment. Pre-processing of the linked graph is first performed to remove noise from the set of original road networks of urban traffic data. Outlier detection strategy is used to efficiently explore the road network and remove irrelevant patterns and noise. The resulting graph is then implemented to train an extended graph convolutional neural network to estimate the traffic flow in the city. To accurately tune the hyperparameter values of the proposed framework, a new optimization technique is developed based on branch and bound. For comparison, an intensive evaluation is conducted with multiple datasets and baseline methods. The results show that the proposed framework outperforms the baseline solutions, especially when the number of nodes in the graph is large. •A unique filtering method based on outlier detection is utilized.•An extended graph convolution neural network is applied to forecast the traffic flow.•A new optimization technique based on branch and bound is presented.