A Data-Driven Approach to Predict Fatigue in Exercise Based on Motion Data from Wearable Sensors or Force Plate.

Area of Science:

• Biomechanics
• Sports Science
• Machine Learning

Background:

• Fatigue significantly increases injury risk in sports training and rehabilitation.
• Early fatigue detection is crucial for adapting training to prevent overtraining and injuries.

Purpose of the Study:

• To develop a data-driven model for automatic fatigue onset prediction and quantification.
• To evaluate the efficacy of force plates (FP) and inertial measurement units (IMUs) in fatigue detection.
• To explore the use of random forest (RF) and convolutional neural network (CNN) regression models for continuous fatigue estimation.

Main Methods:

• Movement data from squats, high knee jacks, and corkscrew toe-touches were captured using FPs and IMUs.
• RF and CNN regression models were employed to estimate participant-specific fatigue levels.
• Analysis focused on the correlation between predicted and self-reported fatigue levels, and the displacement of the center of pressure (COP).

Main Results:

• High correlations (up to 94%) were observed between predicted and self-reported fatigue levels across exercises.
• CNN models achieved the best performance, with FP and IMU data yielding similar results.
• The displacement of the center of pressure (COP) showed a notable correlation with fatigue.

Conclusions:

• The proposed deep neural network model effectively detects subtle changes in motion data for continuous fatigue quantification.
• The methodology shows potential for broad application across various exercises to adapt training and prevent fatigue-related injuries.
• This approach contributes to human motion recognition and personalized exercise program adaptation.