A Graph Neural Network Approach to Citywide Cycling Infrastructure Prioritization

Increasing the share of cycling in urban areas has many benefits. Cycling improves the quality of life by reducing noise and air pollution, improves individual health and, most importantly, reduces greenhouse gas emissions. The most important lever for promoting cycling is the improvement of cycling infrastructure. However, public opinion is often deeply divided, especially when it comes to allocating scarce resources, such as funding and space, between motorized traffic and cycling. Bridging this divide requires a nuanced approach to synthesizing evidence-based evidence for policymakers. To this end, we seek to provide decision-makers with accurate citywide estimates of bicycle volumes that indicate areas where prioritized bicycle infrastructure improvements would be most beneficial.
The city of Berlin currently monitors bicycle movement at a mere 20 locations. We extrapolate cycling volumes from these locations to the entire cityscape at the street level, using newly available data and state-of-the-art machine learning algorithms. In addition to infrastructure and weather indicators, we use data from Strava, a crowdsourced platform that allows users to track cycling activity. Across multiple publications, Strava data is a promising indicator of overall cycling and is particularly useful because it provides street-level information. We combine this data with graph neural networks. This machine learning technique is particularly suited for extrapolating motorized traffic volumes, as it is adept at capturing spatial and temporal dependencies and thereby captures the dependency dynamics of traffic data ideally.
Our research is the first foray into harnessing the computational power of graph neural networks to extrapolate citywide bicycle volumes. Our work is currently still at an early stage, but the positive initial results suggest that leveraging graph neural networks for estimating citywide bicycle volumes outperforms various baselines and has the potential to significantly inform evidence-based decisions on where to prioritize cycling infrastructure improvements in urban areas.