Characterizing the critical role of older people's overall satisfaction with green spaces for their well-being using machine learning methods: Feature extraction and predictive modeling.

This study establishes an integrative machine learning (ML) framework that bridges environmental psychology and data science to investigate the psychological well-being of older adults in urban green spaces (UGS). We applied decision tree (DT), random forest (RF), and artificial neural network (ANN) algorithms, complemented by SHAP interpretability, to multi-dimensional data from 536 seniors in Nanjing, China, aiming to identify key predictors of well-being. While DT achieved the highest accuracy (92.19 %), RF's ensemble approach (87.07 % accuracy) demonstrated superior robustness by effectively mitigating overfitting. Crucially, all models converged in identifying overall UGS satisfaction, a core subjective perceptual metric, as the paramount predictor, underscoring its primacy over traditional accessibility-centric paradigms. SHAP analysis further decoded this global satisfaction into actionable, psychologically salient elements, revealing nonlinear thresholds: wetland parks yielded significant well-being gains (ΔWOOP ≥3.5) only with frequent visits exceeding three weekly and high satisfactions of at least 4 out of 5, while safety facilities and vegetation diversity were identified as key design levers. Our methodology offers a replicable pipeline that balances predictive performance with psychological interpretability. These findings reposition UGS as scalable public health infrastructures for aging well, providing evidence-based, perception-centered strategies to enhance mental and emotional health in urban aging populations.