RESUMEN
The identification of fatigue in personal workers in particular environments can be achieved through early warning techniques. In order to prevent excessive fatigue of medical workers staying in infected areas in the early phase of the coronavirus disease pandemic, a system of low-load wearable electrocardiogram (ECG) devices was used as intelligent acquisition terminals to perform a continuous measurement ECG collection. While machine learning (ML) algorithms and heart rate variability (HRV) offer the promise of fatigue detection for many, there is a demand for ever-increasing reliability in this area, especially in real-life activities. This study proposes a random forest-based classification ML model to identify the four categories of fatigue levels in frontline medical workers using HRV. Based on the wavelet transform in ECG signal processing, stationary wavelet transform was applied to eliminate the main perturbation of ECG in the motion state. Feature selection was performed using ReliefF weighting analysis in combination with redundancy analysis to optimize modeling accuracy. The experimental results of the overall fatigue identification achieved an accuracy of 97.9% with an AUC value of 0.99. With the four-category identification model, the accuracy is 85.6%. These results proved that fatigue analysis based on low-load wearable ECG monitoring at low exertion can accurately determine the level of fatigue of caregivers and provide further ideas for researchers working on fatigue identification in special environments. © 2023 Author(s).