Bolin and colleagues convert pedestrian presence into a temporally structured statistical object. Their central insight is that built density and street centrality explain different dimensions of urban movement: density largely governs the magnitude of pedestrian activity, while relative street position helps distribute that activity within neighbourhoods. The paper operationalises this distinction through hour-by-hour pedestrian counts from Amsterdam, London and Stockholm, derived from anonymised Wi-Fi signals, cleaned, scaled against manual counts and aggregated by street segment. A Bayesian functional ANOVA with negative-binomial likelihood models temporal variation and is compared through cross-validation with k-nearest neighbours, random forests, gradient boosting and neural networks. The methodological contribution lies as much in model interpretability as predictive performance: more opaque machine-learning methods do not automatically improve prediction. The wider significance concerns evidence governance. Urban analytics becomes useful not when it maximises computational complexity, but when variables, assumptions, temporal effects and spatial differences remain interpretable enough to support comparison, prediction and planning judgement.
URBANISM MEETS ART