<p>Campus pedestrian environments play a critical role in alleviating student stress through environmental factors. According to Stress Recovery Theory (SRT), visual exposure to natural elements may trigger restorative psychological responses. However, previous studies have rarely compared multiple campuses or comprehensively examined the combined influence of diverse pedestrian-space characteristics. Meanwhile, traditional questionnaire-based assessments are difficult to implement systematically in highly stimulating environments. To address these gaps, we propose a scalable approach combining street-view imagery (SVI) and machine learning to evaluate the restorative quality of campus pedestrian environments. Across four Nanning campuses, we quantified visual features from 1060 street-view images, then paired these with on-site Restoration Outcome Scale (ROS) scores. Finally, multiple regression and random forest models were used to assess linear and nonlinear links to perceived restorativeness, achieving high accuracy (RMSE = 0.078; MSE = 0.049; R² = 0.602). Key findings are threefold. First, campus restorative quality is systematically tied to a distinctive visual signature, with pronounced heterogeneity across universities. Second, while linear regression confirmed the positive influence of natural elements, the more robust machine-learning model identified vegetation cover and sky visibility as the two most important predictors. Third, partial-dependence analysis uncovered a nonlinear pattern: sky visibility consistently enhanced restorative quality, whereas vegetation’s effect became negative beyond a threshold. This challenges the conventional “more-is-better” assumption for campus greening and indicates that spatial configuration and perceived quality of environmental features, rather than sheer quantity, are more decisive for stress recovery. This provides a scalable, empirical method to optimize campus greenspace for student well-being.</p>