Integration of IoT and machine learning for enhancing physiological workload management of tunnel construction workers

Occupational accidents caused by heavy physical labor remain common among construction workers. Although recent studies have attempted to combine wearable devices with artificial intelligence technology for workload monitoring, several limitations remain: Devices are difficult to deploy on‐site, laboratory‐based studies lack on‐site representativeness, results are difficult to implement, and most models lack physiological threshold management. Therefore, this study proposes an end‐to‐end physiological workload prediction framework for tunnel construction sites, which integrates sensing, classification, and regression modeling. First, a real‐time photoplethysmography‐based Internet of Things sensing system was developed, achieving continuous, non‐intrusive monitoring. The signal showed strong correlation ( r > 0.88) and a low MAPE (<5%), compared with electrocardiogram. Second, a machine‐learning model using static worker features was trained to classify workload levels, with the decision tree achieving the best performance (accuracy: 0.7879; F1‐score: 0.7801). Third, a personalized physiological workload regression model based on the percentage heart rate reserve and work duration was constructed, yielding R 2 > 0.98 and root mean squared error/mean absolute error < 5%. The major innovation of this study lies in integrating sensing, classification, and regression from pre‐task prediction to post‐task verification, establishing a full‐process physiological load management framework for high‐risk construction sites such as tunnels. This approach bridges the gap between lab‐based research and practical deployment, offering strong potential for improving worker health management and advancing smart construction site development.