Machine Learning Approach to Model Physical Fatigue during Incremental Exercise among Firefighters

Area of Science:

• Occupational Health and Safety
• Biomedical Engineering
• Machine Learning Applications

Background:

• Physical fatigue poses significant risks to firefighter health and safety, impairing cognitive function and increasing accident likelihood.
• Current methods for monitoring firefighter fatigue, including subjective scales and on-body sensors, have limitations in accuracy and validation.
• Developing advanced predictive models is crucial for effective fatigue management in safety-critical professions.

Purpose of the Study:

• To develop and validate a physical fatigue prediction model for firefighters.
• To integrate cardiorespiratory and thermoregulatory measures with machine learning algorithms.
• To assess the model's performance in distinguishing between four levels of physical fatigue: low, moderate, heavy, and severe.

Main Methods:

• Collected physiological data (heart rate, breathing rate, core temperature) from 24 firefighters during an incremental running protocol.
• Extracted 21 features from physiological variables and participant characteristics.
• Trained and evaluated various supervised machine learning algorithms, including XGBoosted Trees, to predict fatigue levels.

Main Results:

• The XGBoosted Trees algorithm demonstrated the highest performance with an average accuracy of 82%.
• Specific accuracies for fatigue levels were 93% for low fatigue and 86% for severe fatigue.
• Group cross-validation was identified as the most practical method for assessing model performance.

Conclusions:

• Combining multiple physiological measures enhances physical fatigue modeling accuracy.
• The developed model shows promise as a health and safety management tool for firefighters.
• Further research is needed to validate these findings under field conditions.