Hourly forecasting of traffic flow rates using spatial temporal graph neural networks

Area of Science:

• Transportation Engineering
• Data Science
• Urban Planning

Background:

• Traffic congestion is a significant global issue, causing delays and economic losses.
• Post-COVID-19, understanding evolving transportation patterns is crucial for policymakers.
• Resuming travel necessitates advanced tools for analyzing urban mobility.

Purpose of the Study:

• To evaluate the effectiveness of a Spatial Temporal Graph Neural Network (STGNN) for traffic flow forecasting.
• To analyze traffic patterns in Amsterdam using real-world sensor data.
• To compare STGNN performance against a baseline model for urban traffic prediction.

Main Methods:

• Utilized a Spatial Temporal Graph Neural Network (STGNN) model.
• Trained the model on hourly traffic flow data from 34 sensors in Amsterdam.
• Compared STGNN predictions against a seasonal naive baseline model over a three-month period.

Main Results:

• The STGNN model did not consistently outperform the seasonal naive baseline across all sensors.
• STGNN demonstrated superior performance for traffic sensors situated in close proximity within the road network.
• Performance improvements were observed for spatially clustered sensors.

Conclusions:

• STGNN shows potential for traffic flow prediction, particularly in dense road networks.
• The model's effectiveness is influenced by the spatial relationships between traffic sensors.
• Further research could explore hybrid models or network configurations to enhance STGNN performance.