Stress Detection from Surface Electromyography using Convolutional Neural Networks.

Area of Science:

• Physiology
• Computer Science
• Machine Learning

Background:

• Stress detection is crucial across scientific fields, with increasing interest in automated systems using physiological markers.
• Physiological signals offer reliable and accurate data for stress identification and modeling.
• Machine learning techniques enhance the capabilities of stress detection systems.

Purpose of the Study:

• To explore the application of Convolutional Neural Networks (CNN) for stress detection.
• To utilize surface electromyography (sEMG) signals from the trapezius muscle for stress analysis.
• To evaluate a CNN model that processes raw sEMG signals without pre-computed features.

Main Methods:

• Development and application of a CNN model for analyzing sEMG signals.
• Direct use of sEMG signals, bypassing traditional feature extraction methods.
• Implementation of both multi-class and bi-class classification tasks for stress detection.

Main Results:

• The proposed CNN model demonstrated effective stress detection capabilities.
• Achieved a 73% f1-score for multi-class stress classification.
• Attained an 82% f1-score for bi-class stress classification.

Conclusions:

• CNNs are a viable and effective tool for stress detection using sEMG signals.
• The model's ability to use raw sEMG data offers an advantage over classical machine learning approaches.
• The findings support the potential of automated sEMG-based stress detection systems.