A Data-Driven Approach to Physical Fatigue Management Using Wearable Sensors to Classify Four Diagnostic Fatigue...

Area of Science:

• Rehabilitation Engineering
• Biomedical Data Science
• Human-Computer Interaction

Background:

• Physical exercise is crucial for rehabilitation, but optimal intensity and fatigue monitoring remain challenging.
• Social robots can assist rehabilitation, yet understanding individual fatigue is key to preventing complications.
• Current machine learning fatigue management lacks insight into performance deterioration across diverse conditions.

Purpose of the Study:

• To establish a data analytic framework for fatigue management in walking tasks.
• To define criteria for feature and machine learning algorithm selection for fatigue classification.
• To classify four distinct fatigue diagnosis states.

Main Methods:

• Developed a framework for feature and machine learning algorithm selection.
• Implemented a machine learning classifier to diagnose four fatigue states.
• Evaluated model performance using Inertial Measurement Units (IMUs) and analyzed sensor reduction impact.

Main Results:

• The random forest model achieved high accuracy (≥98%) and F-score (≥93%) using ≤16 features.
• Performance remained robust (≥88%) even with reduced sensor input (one or two IMUs).
• The framework provides a foundation for data-driven fatigue management in rehabilitation.

Conclusions:

• A data analytic approach effectively manages exercise fatigue in rehabilitation.
• Machine learning, particularly the random forest model, shows high potential for accurate fatigue state classification.
• The proposed framework and findings support the development of safer and more effective robot-assisted rehabilitation programs.