From explanations to predictions. Developing a predictive model of pedestrian flows on existing and planned streets.

Paper in proceeding, 2024

Promoting sustainable mobility and urban development hinges on understanding and forecasting pedestrian movement. While empirical studies, not least in Space Syntax research, have used explanatory statistical models to identify spatial parameters influencing pedestrian movement, these models face limitations in forecasting pedestrian flows in future or data-scarce areas. Thus, they are not as useful for scenario analysis, assessment and decision making in urban design and planning. Pedestrian route-choice models are equally challenging as they are highly data demanding and depend on predictors too detailed for early design and planning stages. Instead of complex models, this study proposes a parsimonious predictive model based on street network modelling and a few spatial predictors that can be easily defined and calculated during early project stages.
The paper outlines the methodology and results of the model, which employs LASSO regression in machine learning to predict numbers of pedestrians at the street segment level. The model is trained using data gathered in Stockholm and is first tested by predicting full-day pedestrian counts at street segments of central Gothenburg. The model is evaluated both in relation to predicting the absolute number of pedestrians and their relative distribution within the area. This concise yet effective model shows promising results for early forecast of pedestrian flows in development plans and infrastructural changes and can offer a valuable tool for planners and designers to influence and optimise the distribution of pedestrian flows in various urban contexts.