Development of a Real-Time Stress Detection System

Abstract

This research presents the development of a real-time stress detection system based on physiological signals, with a particular focus on Photoplethysmography (PPG) signals derived from the WESAD dataset. The goal is to accurately classify stress levels using a combination of traditional machine learning models (SVM, Random Forest, XGBoost) and deep learning approaches (CNN, LSTM, and hybrid architectures). The system involves multiple stages, including signal preprocessing, statistical feature extraction, data augmentation, and model optimization. Experimental results show that the LSTM model achieved the best performance, reaching a classification accuracy of over 93.23% and a macro F1-score of 0.92, while maintaining a low prediction latency of approximately 0.06 seconds per sample. These results confirm the suitability of the proposed system for real-time applications. Overall, this method provides a foundation for non-invasive, continuous stress monitoring in real-world settings.