Abstract
Physical exertion led fatigue is a serious threat to occupational health and safety of construction workers worldwide. Its acute effects include a decrease in cognitive abilities, productivity and heightened risk of accidents whereas prolonged physical exertion led fatigue could lead to psychological issues and development of musculoskeletal disorders. To monitor physical exertion, traditionally questionnaires have been used while recent advances have focused on onsite and on-body sensors to automate the process. Considering the limitation of the recent approaches, this study explored the use of combined cardiorespiratory and thermoregulatory measures to model physical exertion using machine learning algorithms. Controlled manual material handling experiments were conducted during a preliminary study to induce exertion at a steady rate involving ten participants. The results revealed that the proposed methodology could predict exertion levels with a high accuracy of 95.3% for combined data modeling of all participants. However, for some predictions, the error between predicted and actual exertion was up to five levels on the Borg-20 scale. To mitigate this issue, individualized machine learning models were used that effectively reduced the maximum error to one level with an average accuracy of 96.7% while using only one-tenth of the total data set. Overall, this study highlights the advantage of using multiple physiological measures for enhancing physical exertion modeling. Notably, the study underpins the use of individualized models for exertion monitoring and management to prevent physical fatigue development and its ill effects.