Introducing WEST GCN-LSTM: A New AI Approach to Predict Urban Mobility
Traffic is more than just a daily nuisance—it’s a major challenge for the design and management of smart, livable cities. As urban areas grow and mobility patterns become more complex, predicting how people and vehicles move across different regions is becoming increasingly vital for everything from public transportation to emergency services.
That’s why researchers from Harokopio University of Athens, the University of Vienna, and the National Technical University of Athens developed a new artificial intelligence (AI) model for regional traffic forecasting. The model, called WEST GCN-LSTM (WEighted STacked Graph Convolutional Network with Long Short-Term Memory), was recently published in the International Journal of Information Management Data Insights. It offers a powerful way to forecast traffic flows by combining advanced graph-based modeling with time-series prediction.
Why Traditional Models Fall Short
Standard traffic forecasting models often treat regions as isolated grids or only consider one kind of information—either where traffic is happening (spatial data) or when it happens (temporal data). But real-world traffic is dynamic and interconnected: what happens in one neighborhood can ripple through an entire city, and those patterns shift over time.
WEST GCN-LSTM: What Makes It Different
The WEST GCN-LSTM model stands out because it brings together spatial and temporal information in a much more integrated way than previous models. Here’s how:
- Shared Borders Policy: The model accounts for how much border two regions share, assuming that longer shared borders are more likely to see movement between them.
- Adjustable Hops Policy: It also adjusts how far information spreads between regions based on how fast different populations (like cyclists or pedestrians) move, allowing it to better reflect real-world mobility.
By combining these insights, WEST GCN-LSTM doesn’t just forecast traffic—it understands it in context.
Real-World Testing
To test the model, the team used a mix of real and simulated data, including pedestrian traffic in New York’s Central Park and long-term cycling data from Berlin. The results were clear: WEST GCN-LSTM outperformed 18 other models, showing significant improvements in accuracy, flexibility, and robustness across different urban scenarios.
Built with EU Support
This research is part of the MUSIT project—an interdisciplinary effort funded by the European Union’s Horizon 2020 programme (Grant Agreement No. 101182585). MUSIT explores how to improve mobility predictions by using AI, advanced sensors, and innovative data modeling techniques.
Towards a More Predictive Urban Future
By delivering more accurate traffic forecasts, WEST GCN-LSTM can help city planners, policymakers, and mobility providers make smarter decisions. From reducing congestion and pollution to improving public transport and emergency response, smarter forecasting is a key building block for the cities of tomorrow.
📖 Read the full paper: Elsevier – WEST GCN-LSTM: Weighted stacked spatio-temporal graph neural networks for regional traffic forecasting