Leveraging Strava Metro Data to enhance urban cycling infrastructure development in Brussels

• We train a neural network on cycling data in Brussels to predict daily cycling counts. • We identify locations with high cycling volumes but insufficient infrastructure. • We differentiate between ‘missing links’ and ‘independent high-demand links.’ Cycling has seen a rising popularity over the past decade, booming even more since the COVID-19 pandemic. To accommodate this increase, appropriate infrastructure needs to be developed, requiring knowledge of traffic volumes, which can be challenging for active mobility. For cycling, one can overcome this through the use of automated or manual bike counters, but these counts are often only performed at strategic locations throughout cities. An additional interesting source of information can come from crowdsourced data which can fill the spatial gaps of counters. Our work explores how Strava Metro data can be used to estimate daily cyclist volumes in Brussels, Belgium, and guide new infrastructure development. We train a neural network to learn the relationship between Strava edge activity and automated counts, considering counter location within the region. We further augment the input Strava data with additional features, such as weather or day of the week. The trained model is then tasked to predict counts of a hypothetical counter on high-volume Strava edges without separated cycling infrastructure. This allows us to evaluate cycling volumes on these edges on equal footing with automated counter volumes. Our analysis identifies two categories of infrastructure gaps—(i) “missing links” where short discontinuities interrupt otherwise protected corridors, and (ii) “independent high-demand links ” without separated cycling infrastructure. Our analysis and methodology offers a reproducible and robust approach to provide insights into cycling flows in cities, allowing us to identify locations for new cycling infrastructure.