A lightweight St-CNN architecture based on deep learning for stress level detection from human physical activities

Area of Science:

• Computational neuroscience
• Machine learning applications
• Wearable technology and health monitoring

Background:

• Stress significantly impacts daily and professional life, necessitating effective management and monitoring strategies.
• Accurate and objective stress detection remains a significant challenge despite various developed methods.
• Physical activity data offers a promising avenue for non-invasive stress level assessment.

Purpose of the Study:

• To propose a lightweight Stress Convolutional Neural Network (St-CNN) architecture for individual stress level detection.
• To evaluate the St-CNN model's performance using the Stress-Lysis dataset.
• To demonstrate the superiority of the St-CNN over traditional machine learning methods in stress detection.

Main Methods:

• Developed a streamlined St-CNN architecture with two fully connected layers, ReLU, and normalization layers for low computational cost.
• Utilized the Stress-Lysis dataset (2,001 samples) including features like body temperature, humidity, and step count.
• Performed 10-fold cross-validation to validate the model's robustness and generalizability.

Main Results:

• The St-CNN model achieved a perfect accuracy rate of 100% on the Stress-Lysis dataset.
• Outperformed traditional machine learning methods such as Decision Tree (DT), K-Nearest Neighbors (KNN), and Support Vector Machine (SVM).
• Achieved 99.85% accuracy with 95% confidence intervals during 10-fold cross-validation, surpassing state-of-the-art approaches.

Conclusions:

• The proposed St-CNN architecture offers a practical and efficient solution for real-time stress monitoring.
• The model's lightweight design is suitable for edge computing environments with minimal computational overhead.
• The St-CNN demonstrates high classification accuracy and superior performance for stress level detection.