Generalized model for mapping bicycle ridership with crowdsourced data

Fitness apps, such as Strava, are a growing source of data for mapping bicycling ridership, due to large samples and high resolution. To overcome bias introduced by generated from only fitness app users, researchers build statistical models that predict total integrating Strava with official counts geographic data. However, studies conducted on single cities provide limited insight best practices modeling generalizability is difficult assess. Our goal develop generalized approach ridership using In doing so we enable detailed more inclusive all bicyclists will support equitable decision-making across cities. We used data, counts, model Average Annual Daily Bicycling (AADB) in five cities: Boulder, Ottawa, Phoenix, San Francisco, Victoria. Using machine learning approach, LASSO, identify variables important predicting cities, independently each city. the LASSO-selected predictors Poisson regression, built city-specific compared accuracy. results indicate prediction road segment concert should include following variables: number riders, percentage trips categorized commuting, safety, income. Inclusion increased performance, R2 ranged 0.08–0.80 0.68–0.92, respectively. accuracy was influenced most count training. For results, capture diverse street conditions, including low areas. Counts collected continuously over long time period, rather than at peak periods, may also improve modeling. Modeling enables better able decision-making.