Understanding how built environment (BE) features influence cyclists’ perceptions is critical for designing human-centric and bike-friendly spaces. While previous research has largely focused on objective cycling conditions, this study addresses a key gap by investigating the non-linear relationships between BE features and cyclists’ subjective perceptions of safety and comfort, through a novel integration of Street View Images (SVIs), computer vision (CV), and explainable machine learning (ML). Using a pairwise voting survey, 150 participants evaluated 300 SVIs, with the results extrapolated to 5,176 locations using CV and ML models. The Shapley Additive Explanations (SHAP) framework quantified threshold effects and highlighted asymmetries between perceived safety and comfort, revealing three key insights: (1) Over 80% of subjective perceptions are explained by spatial chaos, greenery, cycle lane type, feature entropy, perceived road width, and crowdedness, though their relative importance and thresholds vary; (2) Several BE features exhibit threshold effects, including notable inverted “V-shaped” trends, with safety demonstrating higher sensitivity and lower tolerance to BE features compared to comfort; (3) Comfort-oriented interventions are particularly effective in heritage areas where structural constraints limit safety improvements. These findings emphasize the need for localized and context-sensitive strategies that differentiate between safety-oriented and comfort-prioritized interventions, enabling more effective cyclist-friendly urban design.
CORE (COnnecting REpositories)
CORE (COnnecting REpositories)