How human-perceived urban green spaces contribute to mental well-being: a machine learning approach

Global urbanization has heightened attention to urban green spaces (UGS) and their role in supporting mental well-being (MWB). While macro-scale landscape ecology has guided the quantity and morphology of UGS, a human-perception perspective is still needed to clarify how micro-scale features affect MWB and to support higher-quality, fine-grained UGS construction. To address unresolved questions about which UGS perceptual features are critical for MWB, how their nonlinear association trajectories unfold, and how they interact-and given the limitations of traditional statistical methods in automatically detecting complex nonlinear relationships and interactions-this study, grounded in Landsenses Ecology, systematically identifies the micro and essential environmental factors that may affect MWB. The study further applies ecosystem service supply-demand theory to assess how resident-perceived supply-demand matching relates to MWB. Using Beijing Olympic Forest Park as a representative case, an explainable machine-learning approach coupling Random Forest regression with SHapley Additive exPlanations (SHAP) is employed to identify key factors and characterize univariate nonlinear association patterns and pairwise interaction patterns. The results show that: (1) Natural attributes of UGS dominate the model-attributed contributions to MWB, with a relative contribution of 58 %, particularly Amplitude of Topographic Relief, Diversity of Terrain, Diversity of Animal Sounds, Area of Waterbody, and Degree of Greening, whereas demographic variables contribute little, accounting for only 1.1 %. (2) For key factors, supply-demand matching shows an overall positive, nonlinear association with MWB and further exhibits two marginal-pattern types: post-threshold accelerating and post-threshold decelerating. (3) Main effects of key UGS factors are markedly stronger than pairwise interaction contributions. Nevertheless, some factor pairs exhibit trade-offs or synergies under specific supply-demand matching conditions. Overall, the study maps how multi-sensory perceptions of UGS features are associated with MWB, offering scholars a higher-level, integrative vantage that guides the prioritization of future research and experimental design; concurrently, it provides evidence to optimize ecological resource allocation and to deliver high-quality UGS that enhance MWB.