Detecting-Human-Stress-with-the-WESAD-Dataset-From-Signal-to-Prediction

WESAD-Chest Signals Analysis

This project analyzes the WESAD dataset’s chest-worn RespiBAN device data to detect and classify four affective states: Neutral, Stress, amusement and Meditation

About this Dataset

WESAD (Wearable Stress and Affect Detection) is a publicly-available multimodal dataset designed for wearable stress and emotion research. It was recorded in a lab study with 15 participants using both a wrist-worn and a chest-worn device.

Sensor Modalities

• Blood Volume Pulse (BVP)
• Electrocardiogram (ECG)
• Electrodermal Activity (EDA)
• Electromyogram (EMG)
• Respiration (RESP)
• Body Temperature (TEMP)
• Three-axis Acceleration (ACC)

Affective States

1. Baseline (Neutral) – 20 min reading magazines
2. Stress – Trier Social Stress Test (public speaking + mental math)
3. Amusement – Watching humorous video clips
4. Meditation(controlled breathing exercises)

Key Facts

• Chest device sampling rate: 700 Hz (ECG, EDA, EMG, RESP, TEMP, ACC)
• Wrist device sampling rates vary by channel
• Self-report questionnaires accompany each session
• Benchmark performance:
  • 3-class (neutral vs. stress vs. amusement): up to 80% accuracy
  • 2-class (stress vs. non-stress): up to 93% accuracy

Citation

If you use WESAD in your work, please cite:

Schmidt, P., Reiss, A., Duerichen, R., Marberger, C., & van Laerhoven, K. (2018).
Introducing WESAD, a multimodal dataset for wearable stress and affect detection.
Proceedings of the 20th ACM International Conference on Multimodal Interaction (ICMI ’18), 400–408.
https://doi.org/10.1145/3242969.3242985

Disclaimer

You may use this data for scientific, non-commercial purposes only, provided that you give appropriate credit to the original authors. All rights reserved by the original creators.

• Device: RespiBAN (chest)
• Signals used:
  • ECG (heart activity, 700 Hz)
  • EDA (skin conductance, 700 Hz)
  • RESP (respiration, 700 Hz)
• Conditions:
  1. Baseline (reading magazines)
  2. Stress (public speaking + mental math)
  3. Amusement (funny videos)
  4. Meditation (controlled breathing exercise)

Methodology

1. Preprocessing

   • Drop all rows with temp == 0
   • Biosppy filters on ECG, EDA, Resp

2. Window Segmentation
   I fixed each window to 60 seconds (42 000 samples at 700 Hz), and generated three separate segmentations by shifting the window start by:

   • 10 s (7 000 samples)
   • 20 s (14 000 samples)
   • 30 s (21 000 samples)

   Each segmentation produces its own feature table.

3. Feature Extraction
   Per window, compute:

   • Time-domain stats: mean, std, median, min, max, skew, kurtosis, Q1/Q3
   • PSD statistical analysis
   • HR/HRV interpolated features via R-peak detection

4. Oversampling and Scaling

   • Oversampler: SMOTE
   • Scaler : Standard Scaler (Z-score Normalization)

5. Modeling

   • Classifier: Logistic Regression
   • Train/test split: 80/20 stratified by label

Results

I evaluated Logistic Regression using 60 s windows with three different step sizes. Here are the binary (stress vs. non-stress) and multi-class (baseline vs. stress vs. amusement vs meditation) accuracies:

Window Size	Step Size	Binary Accuracy	Multi-class Accuracy
60 sec	30 sec	97%	80%
60 sec	20 sec	95%	81%
60 sec	10 sec	95%	81%

Useful Resources

• GitHub: Stress & Affect Detection example
  https://github.com/jaganjag/stress_affect_detection
• GitHub: Springboard WESAD project
  https://github.com/arsen-movsesyan/springboard_WESAD
• Guide to Electrodermal Activity (University of Birmingham PDF)
  https://www.birmingham.ac.uk/Documents/college-les/psych/saal/guide-electrodermal-activity.pdf
• Choi et al. “Development and evaluation of an ambulatory stress monitor…”
  http://research.cs.tamu.edu/prism/publications/choi2011ambulatoryStressMonitor.pdf

---

References

• Schmidt, P., Reiss, A., Duerichen, R., Marberger, C., & Van Laerhoven, K. (2018).
  Introducing WESAD, a multimodal dataset for wearable stress and affect detection.
  ICMI 2018, 400–408. https://doi.org/10.1145/3242969.3242985
• Healey, J. A., & Picard, R. W. (2005). Detecting stress during real-world driving tasks using physiological sensors. IEEE Transactions on Intelligent Transportation Systems, 6(2), 156–166.
• “From lab to real-life: A three-stage validation of wearable technology for stress monitoring.” MethodsX (2025). https://doi.org/10.1016/j.mex.2025.103205
• Measuring mental workload using physiological measures: A systematic review https://doi.org/10.1016/j.apergo.2018.08.028