Machine learning predicts pedestrian counts from city map features
Estimating Pedestrian Volumes from GIS-Derived Built-Environment Features: A Machine Learning Framework
Machine Learning
Summary
Transportation planners need to know how many people walk at different intersections to make roads safer, but counting pedestrians manually is hard and expensive. The authors created a computer model that uses information from maps about streets and buildings to predict how many people walk at certain times in Portland. Their model does better than the standard method, cutting down prediction errors by around 12 to 19 percent. This helps make better decisions about where to focus on pedestrian safety without needing lots of manual counting.
What this means in practice
- •For urban planners: Estimate pedestrian traffic at intersections from map and land-use data to prioritize safety improvements without costly manual counts.
- •For traffic management agencies: Generate better pedestrian volume predictions for urban road networks to support planning and allocation of resources more efficiently.
- •For smart city software developers: Incorporate built-environment based pedestrian flow predictions to enhance urban mobility platforms and pedestrian safety features.$Commercial implications: Enables new predictive traffic management products that can be sold to municipalities for improved pedestrian monitoring and planning.
Authors
Bahareh Golchin, Banafsheh Rekabdar, Sirisha Kothuri, Joseph Broach
Abstract
Transportation agencies need pedestrian volume estimates across entire road networks to prioritize safety investments, yet manual counts are expensive and cover only a small share of intersections. We present a machine learning pipeline that predicts 2-hour PM peak pedestrian volume at 101 urban intersections in Portland, Oregon, from built-environment, land-use, and street-network features drawn from open GIS data. Starting from the Negative Binomial GLM used in practice, we add feature selection, count-aware gradient boosting, and repeated cross-validation, selecting one configuration by a combined rank over RMSE, MAPE, and SMAPE across four cross-validation strategies. The winner, a histogram-based gradient boosting model with Poisson loss and L1 Lasso feature selection, reduces cross-validated RMSE by 12% over the GLM baseline (89.8 to 78.7) and holdout RMSE by 19% (108.0 to 87.9). Code is released on GitHub.