Combined linear and nonlinear heart rate variability features achieved 93.5% classification accuracy for identifying physical fatigue levels in construction workers.
Observational (n=15)
Can linear and nonlinear HRV analyses combined with machine learning accurately detect and classify physical fatigue in construction workers?
Linear and nonlinear HRV parameters collected via wearable sensors can accurately classify physical fatigue levels in construction workers using machine learning.
Several studies have analyzed heart rate variability (HRV) using nonlinear methods, such as approximate entropy, the largest Lyapunov exponent, and correlation dimension in patients with cardiovascular disorders. However, few studies have used nonlinear methods to analyze HRV in order to determine the level of physical fatigue experienced by construction workers. As a result, to identify and categorize physical fatigue in construction workers, the current study examined the linear and nonlinear approaches of HRV analysis. Fifteen healthy construction workers (mean age, 33.2±6.9 years) were selected for this study. A textile-based wearable sensor monitored each participant’s HRV after they completed 60 min of bar bending and fixing tasks. At baseline, 15, 30, 45, and 60 min into the task, participants were given the Borg-20 to measure their subjective levels of physical fatigue. Nonlinear e.g., R-R interval (RRI) variability, entropy, detrended fluctuation analysis and linear (e.g., time- and frequency-domain) HRV parameters were extracted. Five machine learning classifiers were used to identify and discern different physical fatigue levels. The accuracy and validity of the classifier models were evaluated using 10-fold cross-validation. The classification models were developed by either combining or individualized HRV features derived from linear and nonlinear HRV analyses. In the individualized feature sets, time-domain features had the highest classification accuracy (92%) based on the random forest (RF) classifier. The combined feature (i.e., the time-domain and nonlinear features) sets showed the highest classification accuracy (93.5%) using the RF classifier. In conclusion, this study showed that both linear and nonlinear HRV analyses can be used to detect and classify physical fatigue in construction workers. This research offers important contributions to the industry by analyzing the variations in linear and nonlinear HRV parameters in response to construction tasks. This study demonstrates that HRV values changed significantly in response to physical work, indicating a change in the relative activity of cardiac autonomic functions as a result of fatigue. Using the ways in which HRV parameters vary in response to increased workloads provides a sensitive marker for contrasting construction workers with and without cardiovascular disease. It also allows the site manager to track how quickly workers fatigue, so that they can switch up their workload to reduce the likelihood that any one worker would get severely exhausted, or to suggest that workers who are already severely fatigued take a break to prevent further injury.
Anwer et al. (Sat,) conducted a observational in Physical fatigue (n=15). Linear and nonlinear HRV analysis was evaluated on Classification accuracy of physical fatigue levels. Combined linear and nonlinear heart rate variability features achieved 93.5% classification accuracy for identifying physical fatigue levels in construction workers.