Fatigue Identification Models for Firefighters Based on Electrocardiogram Signals.

Abstract

To identify physiological indicators and develop effective models for recognizing training-induced physical fatigue in firefighters. Heart rate variability was analyzed in eighteen firefighters before and after physical training. Electrocardiogram signals were processed using linear and nonlinear methods. Feature dimensionality was reduced using two selection approaches, and fatigue recognition models were developed using multiple machine learning and deep learning algorithms. Physical fatigue increased mean heart rate and the low frequency to high frequency ratio, while eight indicators, including mean R-R interval, short and long term variability, and sample entropy showed significant decreases. The least absolute shrinkage and selection operator based one-dimensional convolutional neural network achieved the highest classification accuracy of 94.64%. Electrocardiogram-derived indicators provide a feasible and objective approach for identifying physical fatigue in firefighters, supporting occupational fatigue monitoring and safety management.