<p>Understanding urban housing price dynamics and their underlying mechanisms is crucial for designing effective urban economic policies and spatial governance strategies. While existing research has predominantly focused on macro-level features of the built environment, such as land use structure, there remains a limited systematic understanding of the role of perceived environmental quality at the micro scale. Moreover, the nonlinear associations between these factors pose further analytical challenges. This study develops an interpretable machine learning framework that integrates crowdsourced perception data and street-level imagery to empirically examine the nonlinear and interactive effects of complex urban environmental factors on housing price changes. Utilizing XGBoost models in conjunction with SHAP analysis, the findings reveal that both built and perceived environmental variables exhibit pervasive nonlinear relationships with housing price dynamics, with marginal effects showing distinct threshold patterns across variables. Furthermore, strong interactive effects between subjective perceptions and objective environmental features are identified, often surpassing the main effects of individual variables in multiple scenarios. These results underscore the multidimensional and composite nature of spatial housing price responses. The study highlights the necessity of incorporating perceived environmental data into housing market analyses and provides empirical evidence to support the formulation of more resilient and spatially adaptive urban housing policies.</p> Graphical Abstract <p></p>