Classification of exercise fatigue levels by multi-class SVM from ECG and HRV

Area of Science:

• Physiology
• Biomedical Engineering
• Machine Learning

Background:

• Exercise fatigue classification is crucial for optimizing training and preventing overexertion.
• Electrocardiogram (ECG) and heart rate variability (HRV) are valuable physiological signals for monitoring fatigue.
• Current methods for fatigue classification may lack accuracy or real-time applicability.

Purpose of the Study:

• To develop a novel hybrid approach for accurate and timely exercise fatigue classification.
• To combine deep neural network features from ECG images with linear HRV features.
• To evaluate the proposed method's performance on public and self-collected datasets.

Main Methods:

• ECG signals were transformed into 2-D images using Short-Time Fourier Transform (STFT).
• Image features were extracted using the Visual Geometry Group (VGG) deep learning model.
• Combined ECG image features and linear HRV features were fed into various classifiers.

Main Results:

• The hybrid approach achieved high accuracy, sensitivity, and F1-scores on both datasets (e.g., 96.90% accuracy on EPHNOGRAM).
• Concatenated features demonstrated superior classification performance compared to individual feature sets.
• The system's calculation time was significantly reduced, enabling real-time application.

Conclusions:

• The proposed hybrid method effectively classifies exercise fatigue by integrating deep learning and HRV analysis.
• This approach offers improved accuracy and timeliness for real-time exercise fatigue monitoring.
• The findings suggest a promising tool for athletes, coaches, and healthcare professionals.