Predicting Fatigue in High-Intensity Turnaround Operations Using a Machine Learning Approach

Workplace fatigue is a critical issue in high-risk industries such as construction, manufacturing, and mining, as it can reduce workers' readiness (fit to work), endanger their safety, and decrease productivity. This study developed a machine-learning-based model to predict work readiness by utilizing physiological data (Heart Rate (HR), Blood Pressure, VO₂ Max, Oxygen Saturation), subjective data (Multidimensional Fatigue Inventory (MFI-20)), and cognitive data (Psychomotor Vigilance Test (PVT-3)). The data was collected from participants performing intensive activities during the Turnaround (TA) period. The analysis results showed a significant increase in fatigue after 10 hours of work (p = 0.000; R² = 91.65), indicating a strong influence of work duration on fatigue. The Random Forest (RF) model demonstrated the best performance in predicting MFI scores (R² = 0.9989; Root Mean Square Error (RMSE) = 0.064616) and showed adequate accuracy for HR (R² = 0.7996; RMSE = 127.10849). Among the various physiological parameters analyzed, HR and the MFI-20 were the most representative in predicting work-related fatigue. Therefore, for field implementation, measurements can be focused on these two indicators to obtain efficient results without reducing the prediction accuracy. These findings demonstrate that machine learning is effective for predicting fatigue and supports strategic decision-making in occupational safety risk management.