Non-linear associations between the built environment and outdoor activity duration: An application of gradient boosting decision trees

Outdoor activities benefit human physical and mental health. However, how the built environment influences the duration of location-based outdoor activities is unclear. To address this gap, we measured the built environment using different methods including Geographic Information Systems and a semantic segmentation technique with Google Street View images and associated these independent variables with outdoor activity duration by employing a gradient boosting decision tree model. We found neighborhood characteristics play a much more important role in influencing outdoor activity duration than location characteristics (features about a specific area where outdoor activities take place). Although all seventeen variables had a non-linear and threshold impact, distance to the nearest public transport station is the most significant predictor, followed by residential density, distance to the nearest school, and street connectivity. These findings allow urban planners and policymakers to give detailed advice on planning and designing built environments that better promote outdoor activities.