A predictive geospatial engine for city-wide pedestrian-centric urban regeneration
The Geospatial Pedestrian Activity Engine (GPAE) uses a LightGBM model to predict high-resolution, city-wide pedestrian counts across Loughborough, UK. The framework integrates 29 monitoring points of WiFi sensor data and computer vision video counts with spatial built environment features, achieving a Leave-One-Out Cross-Validation performance of R2 = 0.89 and an MAE of 400 pedestrians per day. Feature analysis indicates that pedestrian volume is driven primarily by the joint presence of social amenities, daily necessities, and residential density rather than transit infrastructure alone.
Why it matters — Urban planners can now map continuous pedestrian activity across entire cities without dense physical sensor networks, using those predictions to systematically categorize streets into four intervention typologies for targeted public realm investments.
Caveat: The predictive framework was developed and validated on a single small city using only 29 monitoring points, which may limit immediate generalizability to larger or structurally different urban environments.