Exploring functional zone-dependent nonlinear associations between objective features and subjective perceptions: A case study in Beijing

Urban functional zones shape both the configuration of the physical environment and the ways individuals perceive and interact with urban spaces. However, the nonlinear associations between zone-specific objective features (e.g., greenery, traffic lights) and subjective perceptions (e.g., beauty and boredom) remain underexplored. This study proposes an analytical framework to investigate such associations in Beijing, China, by integrating street view images (SVIs), panoptic segmentation, and machine learning models. Specifically, we extracted nine objective features and six subjective perceptions from SVIs and modeled their nonlinear associations across different functional zones using 36 XGBoost models combined with SHAP analyses. Model performance was robust, with XGBoost achieving the highest average R2 (0.221) and one of the lowest average RMSE values (0.031) compared to four baseline machine learning models, underscoring its generalizability across perception dimensions and functional zone types. The results reveal three key findings. First, road users and safety barriers consistently ranked among the top contributors to subjective perceptions across all functional zones, while elements like traffic signs exhibited zone-specific effects. Second, the nonlinear associations exhibited varying threshold effects across functional zones, highlighting the context-dependent influence of environmental factors; for example, greenery visibility reduced safety perception in residential zones but enhanced it in commercial zones. Third, typical nonlinear patterns, including V-shaped and threshold effects, were observed, reflecting phenomena such as sensory saturation and diminishing marginal utility. These findings contribute to a deeper understanding of the complex dynamics between urban form and human perception, offering insights for adaptive, human-centric urban design strategies.