Developing a Recovery Pattern Model for Construction Workers: A Machine-Learning Approach

Ensuring proper work–rest cycles is a promising strategy for relieving fatigue in construction workers. However, inadequate recovery is common among construction workers due to inappropriate rest arrangements, increasing the risk of excessive fatigue. How to effectively develop rest plans for construction workers is challenging because individual differences in recovery rates can reduce the effectiveness of standardized rest schedules, while fully personalized rest arrangements are rarely feasible on construction sites. This study aims to predict the recovery patterns of construction workers after working to exhaustion, providing a basis for recovery pattern–based management that considers individual differences while remaining practically feasible. Field studies were conducted at five construction sites in mainland China, involving 199 construction workers, between March and July 2024. Personal and physiological variables of the workers, along with environmental parameters on construction sites, were collected. Three fundamental recovery patterns—rapid, moderate, and slow recovery—were identified through time-series clustering analysis. Construction workers with rapid recovery patterns can achieve approximately 95% recovery after a 40-min rest, while those with moderate and slow recovery patterns achieve approximately 73% and 49% recovery, respectively. To determine the recovery pattern of each worker, a machine-learning-based prediction model was developed that comprehensively considered the personal characteristics of workers and the environmental conditions to which they were exposed. The developed model demonstrated good performance, achieving an accuracy of 87% on the external test dataset. This study proposes grouping construction workers by their predicted recovery patterns to implement tailored rest schedules. This approach ensures adequate recovery by balancing individual physiological differences with on-site practical constraints, thereby providing actionable decision support for rest planning.