Deep Learning-Based Fatigue Monitoring in Natural Environments: Multi-Level Fatigue State Classification

Area of Science:

• Biomedical Engineering
• Machine Learning Applications
• Wearable Technology

Background:

• Escalating workloads increase fatigue, impacting health, safety, and productivity.
• Prior fatigue monitoring studies often lack real-world applicability, relying on controlled experiments.
• Assessing daily fatigue in natural environments remains a significant research gap.

Purpose of the Study:

• To develop and validate a daily fatigue monitoring system using wearable ECG.
• To compare machine learning models with a novel deep learning approach for fatigue classification.
• To establish a reliable method for assessing fatigue in everyday life.

Main Methods:

• Recruited 12 subjects for a 14-day monitoring period.
• Collected electrocardiogram (ECG) signals using a wearable device during daily activities.
• Developed and evaluated machine learning models (manual features) and a deep learning model (C-BL) for fatigue classification.

Main Results:

• The C-BL deep learning model achieved an 83.3% accuracy rate in classifying fatigue levels.
• The end-to-end deep learning model demonstrated superior performance compared to other methods.
• The system successfully monitored fatigue in a real-world environment.

Conclusions:

• The developed wearable ECG-based system is reliable for daily fatigue monitoring.
• Deep learning models show significant promise for accurate fatigue assessment in natural settings.
• This approach can contribute to proactive health management and safety interventions.