Urban freight models are crucial in managing city logistics, yet they often fall short in construction transport. Unlike regular freight flows driven by economic activity and consumer demand, construction logistics are shaped by a unique mix of project-specific and local contextual factors. Traditional freight models, which rely on high-level economic indicators, struggle to deal with this complexity—leading to oversimplified or inaccurate representations of construction-related traffic.
A recent proof-of-concept study tackled this issue by testing whether machine learning (ML) could help forecast construction site transport demand using more granular, construction-specific data. The researchers combined widely available data points about construction projects—such as size, location, and timeline—with contextual features, hoping that this mix would improve the ability to predict how many truck trips a site would generate.The results were sobering but instructive.
Even with advanced ML techniques, predicting the number of transports to a construction site proved highly challenging. The main reason? The high variance in the data. Each project is different, and many influencing factors—such as delivery scheduling practices, local traffic restrictions, or on-site storage capacity—are not easily captured in standard datasets. As a result, while the models could identify general patterns, their predictive power for unseen (test) data remained limited.
Interestingly, one aspect that did work was feature selection. By excluding irrelevant or noisy features, the models performed better than when all available data was used indiscriminately. This highlights that while the available data may not be perfect, refining the input yields value. It also underscores the need for more consistent, detailed, and project-specific data collection in the construction sector.
This study’s contribution lies in opening the black box of construction transport demand. It shows that incorporating project-specific variables—rather than relying solely on macroeconomic proxies—can improve our conceptual understanding of this segment of urban logistics. Although the models weren’t accurate enough for forecasting yet, they reveal which data types matter most and should be prioritized for future research and model development.
This work offers two takeaways for urban planners, freight modelers, and construction contractors: first, that construction transport demand cannot be treated like conventional freight, and second, that better, more context-rich data is essential for any serious attempt at predicting its impacts. For now, it’s not about perfect predictions but about building a smarter foundation for them.
Source: Brusselaers, N., Hjorth, S., Fredriksson, A. and Gundlegård, D. (2025), “The potential of machine learning modeling to predict urban construction transport demand”, Smart and Sustainable Built Environment, Vol. ahead-of-print No. ahead-of-print. https://doi.org/10.1108/SASBE-12-2024-0558