Smaller cities can achieve more accurate traffic predictions without deploying large sensor networks. At the Belgian AI Week, imec showed how transfer learning allows AI models trained in data‑rich cities to be reused effectively in cities with limited traffic data.
AI traffic models that work beyond large cities
During the Belgian AI Week on 17 March, imec researchers presented new results demonstrating how transfer learning can significantly improve traffic state predictions in cities with fewer sensors and limited historical data.
Transfer learning allows AI models trained in large, data‑rich cities to be adapted to smaller or less instrumented cities. This approach reduces the need for costly data collection and enables more cities to benefit from mature, high‑performing AI models.
Proven gains with minimal local data
Imec’s experiments show that even with only a small amount of local data, pretrained traffic models can achieve higher prediction accuracy than models trained exclusively on local datasets. In several cases, transfer‑learning‑based models clearly outperformed city‑specific baselines.
These results highlight the strong potential of transfer learning for many mobility and smart‑city applications, supporting more inclusive and scalable AI adoption across cities with varying levels of digital maturity.
From research to real‑world testing thanks to CitCom.ai
The research was presented by Laure De Cock (imec) and Mahdi Rahmimiasl (imec – IDLab Antwerp) at the BOSA offices in Brussels. The work was carried out in collaboration with multiple cities, including City of Mechelen and Brussels Mobility, both CitCom.ai partners and the city of Bruges and Hamburg that provided essential traffic datasets.
Thanks to these collaborations, the team was able to test AI models across diverse urban contexts and sensor densities, increasing the practical relevance and robustness of the research.
Enabling more inclusive urban AI
By sharing these findings during Belgian AI Week, imec aims to accelerate the adoption of transfer‑learning‑based AI solutions for urban challenges. Particularly in smaller or data‑poor cities. Initiatives such as citcom.ai play a crucial role by offering access to varied testbeds and enabling collaboration between cities, researchers and industry partners.
Interested in exploring how transfer learning or CitCom testbeds could support your city or organization?
Get in touch with Laure De Cock at laure.decock@imec.be
