Transfer learning for traffic state predictions in small and medium-sized cities

Abstract

In Flanders, 80% of the population resides in small and medium-sized cities, which face significant challenges in data-driven traffic management due to a lack of data. Although deep learning models surpass traditional methods for traffic state prediction, they require extensive data, often unavailable in these smaller cities. Transfer learning offers a potential solution by utilizing models trained in data-rich environments to enhance predictions in data-scarce regions. This ongoing research investigates the application of transfer learning for traffic state prediction in small and medium-sized cities. Preliminary tests in Greater Manchester and Lisbon under the TANGENT H2020 project have shown promising results. The next phase involves applying this technique to the Flemish city of Mechelen as part of the CitCom.ai project. The goal is to improve traffic state prediction in data-limited environments, thereby contributing to more efficient urban traffic management.