Abstract
Stress significantly affects both daily and professional life, highlighting the need for effective management strategies and continuous monitoring. While many methods have been developed, achieving accurate and objective stress detection remains a key challenge. This study proposes a lightweight Stress Convolutional Neural Network (St-CNN) architecture designed to detect individual stress levels based on physical activity data. The method was evaluated using the publicly available Stress-Lysis dataset, which contains 2,001 samples with features such as body temperature, humidity, and step count. The St-CNN model is composed of two fully connected layers, along with ReLU and normalization layers, forming a streamlined architecture optimized for low computational cost. It achieved a perfect accuracy rate of 100% and outperformed traditional machine learning (ML) methods, including Decision Tree (DT), K-Nearest Neighbors (KNN), and Support Vector Machine (SVM). The model was additionally validated through 10-fold cross-validation with 95% confidence intervals, achieving an accuracy of 99.85%. The proposed method outperformed state-of-the-art approaches on the Stress-Lysis dataset, achieving superior performance with a lightweight architecture for stress level detection. In conclusion, the proposed St-CNN architecture provides a practical and efficient approach for real-time stress monitoring in edge computing environments, combining high classification accuracy with minimal computational overhead.