Classification of Fatigue Phases in Healthy and Diabetic Adults Using Wearable Sensor

Area of Science:

• Biomechanics and Motor Control
• Machine Learning in Healthcare
• Human-Computer Interaction

Background:

• Muscle fatigue, characterized by a loss of force-generating capacity, can exacerbate tremor.
• Quantifying tremor is crucial for early fatigue detection, enabling interventions to prevent injuries and enhance task accuracy.
• Existing methods lack robust systems for real-time fatigue phase classification.

Purpose of the Study:

• To develop and evaluate a system for recognizing voluntary effort and detecting distinct phases of muscle fatigue.
• To classify hand tremor data to differentiate between rest and effort states.
• To distinguish between early and late stages of muscle fatigue.

Main Methods:

• Collected accelerometer data from the wrist and finger of the dominant hand during rest and voluntary effort tasks.
• Extracted time and frequency domain features from tremor signals using window lengths of 45 and 90 samples.
• Applied machine learning classifiers, including decision trees, k-nearest neighbors, support vector machines, and ensemble methods, with 5-fold cross-validation.

Main Results:

• An ensemble classifier achieved 96.1% accuracy in recognizing rest versus effort tasks using a 45-sample window.
• The same ensemble classifier distinguished between early and late fatigue phases with approximately 98% accuracy.
• The random subspace method within the ensemble classifier proved effective for both task classification and fatigue phase detection.

Conclusions:

• Machine learning analysis of hand tremor is a viable method for detecting muscle fatigue and effort states.
• The developed ensemble classifier demonstrates high accuracy in differentiating fatigue phases, offering potential for injury prevention.
• This system can aid in optimizing task performance and managing worker fatigue in physically demanding occupations.