Profiling spatiotemporal contexts associated with compound exposure disadvantages in noise and greenery: A study using mobile sensors and explainable machine learning

Although exposures to noise or greenery in isolation have been widely studied, limited research has explored the nonlinear associations between spatiotemporal contexts and compound disadvantages from a daily mobility perspective. Drawing on data from mobile sensors, GPS, satellite imagery, points of interest, and activity diaries, this study measures real-time noise and greenery exposures along relevant spatiotemporal contexts. Methodologically, daily mobility is delineated through a spatiotemporal threshold approach, modeling them as stay and move event sequences that underpin the development of a compound disadvantage index. A novel framework, integrating GPBoost with residual-based bootstraps and SHAP, subsequently accommodates time-sequence dependencies and uncertainties to investigate nonlinear relationships between spatiotemporal contexts and compound disadvantages. The main findings indicate that (i) the neighborhood effect averaging problem is apparent in noise and greenery measurements; (ii) socioeconomic status coupled with spatiotemporal contexts can infer individual compound disadvantages; (iii) employment status constitutes the foremost socioeconomic predictor of compound disadvantages; and (iv) the bootstrap-SHAP methodology pinpoints spatiotemporal contexts with statistically significant effects on compound disadvantages, concurrently disclosing thresholds and interactions with socioeconomic statuses.