Objective Urban parks, as essential components of urban green spaces, serve as important venues for the elderly, supporting both physical and mental well-being. Compared to other age groups, the elderly show a higher demand for engaging in activities within green spaces, and their perception of these environments plays a key role in shaping their activity patterns. Understanding how these perceptions influence behavior is crucial for addressing the needs of aging populations in rapidly urbanizing societies. While previous studies have examined the influence of green spaces on the elderly, most have emphasized objective environmental characteristics — such as vegetation coverage, facility distribution, and spatial accessibility — while largely overlooking subjective perceptions that often guide behavior more directly. Subjective experiences such as comfort, attachment, and satisfaction with environmental quality may serve as critical mediators between green space features and health-promoting behaviors. To address this gap, this research employs the extreme gradient boosting (XGBoost) model to investigate the nonlinear relationship between the elderly’s perception of urban park green spaces and their activity characteristics, aiming to inform age-friendly park design and evidence-based green space planning strategies.

Methods This research is based on 779 valid questionnaires collected from five representative urban parks in Fuzhou, a city experiencing rapid demographic aging and urban expansion. The independent variable — the elderly’s perception of green spaces — includes 12 key indicators: accessibility, safety, rationality of fitness and recreational facilities, sanitation, green space quality, maintenance, ecological environment, acoustic comfort, air quality, sense of attachment, and sense of place. These factors are measured using a standardized 7-point Likert scale, ensuring comparability and internal consistency. The dependent variable covers four dimensions of activity level: weekly activity frequency, average activity duration, diversity of activity types, and total weekly activity amount. Descriptive statistics and Spearman correlation analyses are conducted using Excel 2013 and SPSS 26.0 to identify initial associations between variables. Based on significant correlations, an interpretable XGBoost machine learning model is constructed to explore variable importance and detect potential nonlinear relationships between perception indicators and activity outcomes.

Results 1) The elderly generally hold positive perceptions of park environments, particularly in terms of air quality and the ecological environment, indicating a high baseline satisfaction with environmental conditions in the research areas. Differences in perception across gender and age subgroups are found to be minimal, suggesting that subjective evaluations of park environments are relatively consistent among demographic segments. In contrast, total activity levels show substantial variation, which may be attributed to individual factors such as physical condition, recreational preferences, social habits, and park accessibility. 2) Among the perception indicators, sense of place and sense of attachment emerges as the strongest predictors of activity frequency, underscoring the psychological dimensions of green space use. Ecological environment and green space quality significantly influence activity duration, reflecting the importance of natural aesthetics and environmental quality in sustaining longer park visits. Air quality is found to be the most influential factor driving activity diversity, while ecological environment has the greatest overall impact on total activity engagement. 3) The XGBoost model reveals significant nonlinear relationships between perception indicators and all four activity measures. Sense of place, ecological environment, and green space quality shows the most robust effects. Notably, the ecological environment exhibits an inverted U-shaped influence, indicating that while moderate improvements in ecological features promote engagement, excessive environmental complexity or overdesign may reduce activity levels beyond a certain threshold. Additionally, air quality and acoustic comfort play important roles in enhancing activity frequency and diversity, highlighting the value of environmental comfort for encouraging outdoor participation among the elderly.

Conclusion This research demonstrates the effectiveness of the XGBoost model in capturing the complex and nonlinear associations between the elderly’s perception of green spaces and their activity behaviors. Key influencing factors include sense of place, ecological environment, air quality, rationality of park facilities, and overall green space quality. The findings show that both insufficient and excessive facility provision may suppress activity participation, pointing to the need for balanced design strategies. These results contribute to a broader theoretical understanding of how subjective environmental perceptions shape behavioral patterns among the elderly. The research also emphasizes the practical relevance of environmental comfort — particularly air and acoustic quality — in facilitating diverse and frequent outdoor activities. Despite limitations in sample coverage and cross-sectional design, this research provides valuable insights and empirical evidence for optimizing the design and renovation of age-friendly parks. It also lays a methodological foundation for future longitudinal and cross-regional studies exploring behavioral responses to green space interventions. Ultimately, this research supports the high-quality development of urban green spaces and advances policy goals centered on healthy aging and sustainable urban living.