Investigated Predictors of Construction Workers’ Ongoing Fatigue: Random Forest Approach

Abstract

The construction industry has a higher rate of occupational injuries due to human error than other industries, primarily because of its labor-intensive nature. Human error is often associated with workers’ ongoing fatigue. Therefore, it is essential to classify and predict fatigue-related factors in detail to prevent human error resulting from fatigue. Although numerous studies aim to identify construction workers’ fatigue, they must be enhanced by incorporating diverse data types and emphasizing onsite application. In this study, we adopted a random forest to develop a machine learning model to classify and predict fatigue levels for construction workers. Using feature importance, we extracted essential factors associated with construction workers’ fatigue and suggested fatigue management strategies. The random forest model achieved an accuracy of 76.5%, identifying the optimal combination of fatigue predictors based on feature importance. This combination included heart rate, work time, work intensity, activity, accelerometer, activity variation, and angular velocity. The proposed fatigue management strategy comprises two steps: Step 1 involves routine management, while Step 2 focuses on intervention. This study investigates fatigue predictors while considering uncertainty and establishes fatigue management strategies for real-world construction sites. Consequently, site managers better understand workers’ health and fatigue, enhancing practical worker management.