Physiological Stress in Elderly Residents of Densely Populated Urban Villages: A Skin Conductance Study with Interpretable Machine Learning Modeling

High-density urban villages pose significant environmental challenges to the aging population. Beyond traditional exposures such as noise and air pollution, older adults may experience heightened physiological stress due to visual exposure within street environments, yet the precise micro-environmental triggers of physiological stress remain poorly understood. This study investigates how street-level visual elements relate to elderly walkers’ physiological stress. We conducted on-site walking experiments and monitored the Skin Conductance Level (SCL) of 81 elderly participants walking through two typical urban villages in Lingnan, China. We used a semantic segmentation algorithm to quantify visual environmental elements from first-person-view images and employed a CatBoost (Categorical Boosting) model to predict stress levels. The explainable model (SHAP, SHapley Additive exPlanations) was then used to interpret the complex relationships. The model demonstrated strong predictive power (e.g., R2 = 0.72). SHAP analysis revealed roads and sidewalks as the most dominant predictors of SCL changes, exhibiting significant non-linear effects. Their influence surpassed that of environmental aesthetics like vegetation, which showed a more complex, at times even negative, association with stress reduction. The presence of buildings also exhibited a stress-reducing effect, though less so than roads and sidewalks. Key findings revealed the following: (1) Foundational walking infrastructure is the primary determinant of physiological well-being for elderly pedestrians in high-density environments. (2) The stress-reducing effects of vegetation are context-dependent, while buildings function as a form of “social infrastructure” in mitigating stress. Our findings provide crucial, evidence-based guidance for prioritizing interventions in age-friendly urban renewal projects. Our framework offers a transferable tool for human-centered environmental assessment.