Spatiotemporal exploration of Melbourne pedestrian demand

Creating a cityscape conducive to a safe and efficient pedestrian experience requires a holistic understanding of the relationship between the built structures of a city and the movement of individuals within it. To empower policy makers to design, implement, and successfully deliver measures aimed at reducing footpath congestion and improving pedestrian safety the link between pedestrian volume and different features of the built environment needs to be investigated. Observed pedestrian counts at 50 intersections across the City of Melbourne are used as the input dependent variable of two regression models. A global ordinary least squares regression model and a local geographically weighted regression model are generated and evaluated for best fit of purpose. Spatiotemporal statistical handling is employed to clean the data of contextual anomalies. The output of the regression models identified eight key features as the most statistically significant predictors of pedestrian volume in Melbourne, Australia. These features include distance to schools and train stations and measures of footpath connectivity. This study reveals that due to significant spatial and temporal non-stationarity exhibited between pedestrian count sensors and built environment variables, the geographically weighted regression is the most appropriate modelling technique. This paper presents a methodology for the creation of a robust pedestrian prediction model.