Enhancing Stress Detection Employing Physiological Signals from the WESAD Dataset: A Machine Learning Approach with SMOTE

Numerous medical disorders, including diabetes, heart disease, and hypertension, are significantly influenced by stress. Physiological markers for real-time stress detection are becoming more popular as wearable health monitoring devices are widely used. This research uses the WESAD (Wearable Stress and Affect Detection) dataset, which contains multimodal physiological data, including ECG, EDA, EMG, respiration, and temperature, to predict stress levels. We apply four machine learning classifiers to this dataset, focusing on addressing the class imbalance using the Synthetic Minority Oversampling Technique (SMOTE). The outcomes illustrate that Decision Trees outperform other classifiers with an accuracy of 96.27%.  For future work, efforts can be directed towards incorporating additional modalities, such as EEG and eye-tracking data, to improve stress detection accuracy further. Longitudinal data collection could also help understand stress over comprehensive periods, providing insights into chronic stress patterns.