A Mobile Regression Framework with Context-Aware Sensing for Real-Time WBGT Forecasting and Heatstroke Risk Assessment

Heatstroke is a critical public health concern intensified by global warming, especially in rural areas, where access to traditional Wet Bulb Globe Temperature (WBGT) devices is limited due to their cost and resource-heavy settings. To address the concern, this study proposed a scalable and cost-effective alternative to physical WBGT devices. It presents a mobile regression framework equipped with context-aware sensing for real-time forecasting of WBGT and heatstroke risk assessment. Using ten years of meteorological data from NASA POWER and OpenWeatherMap, we trained and evaluated four regression models. The random forest model outperformed linear regression, k-nearest neighbors, and decision trees by achieving the highest R² (99.997%) and adjusted R² score, along with the lowest error metrics (MSE, RMSE, MAE, MAPE close to zero). The results are validated through a 10-fold cross-validation, and the model is deployed in HeatMeter, a lightweight mobile application capable of providing real-time WBGT predictions and context-sensitive personalized heat safety recommendations. It tested and demonstrated high responsiveness with an average inference time of 0.0067 seconds. Additionally, we propose a novel use of ambient light sensing on smartphones as a proxy for solar radiation to eliminate hardware dependency. This framework demonstrates a scalable and cost-effective solution for heat risk assessment in low-resource environments.