Physiological Stress Features from Three Open Datasets: HRV, EDA, and Multi-Modal Signals for Stress Detection

This dataset contains pre-extracted physiological features for stress detection, derived from three publicly available open datasets: the Non-EEG Neurological Status dataset (PhysioNet, n=15, 4 stress conditions), the DRIVEDB Stress Recognition in Automobile Drivers dataset (PhysioNet, n=14, 3 conditions), and the Stress Detection in Nurses dataset (Dryad, n=290, 3 stress levels). Features were extracted using 30-second and 60-second sliding windows and cover HRV time-domain, frequency-domain, and non-linear measures, as well as EDA, skin temperature, respiration, and EMG signals where available. Entropy variants (sample entropy) are provided as separate files. Four harmonised cross-dataset combinations are included: a two-dataset combination (Drivers + Non-EEG, 30s and 60s windows) and a three-dataset combination (all sources, 60s), the latter provided in both original imbalanced form and SMOTE-resampled (balanced) form. Feature extraction was performed using the hrvanalysis and neurokit2 Python libraries; extraction and classification scripts with dependencies are included in the code/ folder. The dataset directly supports the analyses reported in the associated publication (Ladakis et al., 2025, Journal of Medical and Biological Engineering, doi: 10.1007/s40846-025-00958-z).