WESAD — Wearable Stress and Affect Detection
Introduced by: Schmidt et al., 2018
DOI: 10.1145/3242969.3242985
Access: UCI Repository — free download, no auth required
Tier: 1
Overview
The most widely used public dataset for wearable stress detection. 15 subjects (2 excluded for device issues → 13 usable). Lab-based protocol with Trier Social Stress Test (TSST).
Signals
| Modality | Device | Signals |
|---|---|---|
| Wrist | Empatica E4 | BVP (64 Hz), EDA (4 Hz), ACC (32 Hz), TEMP (4 Hz) |
| Chest | RespiBAN | ECG (700 Hz), EDA (700 Hz), EMG (700 Hz), RESP (700 Hz), TEMP (700 Hz), ACC (700 Hz) |
Labels
- Baseline (neutral)
- Stress (TSST: public speaking + mental arithmetic)
- Amusement (funny video)
- Meditation (guided breathing)
Baselines
- 3-class (baseline/stress/amusement): RF with wrist data ~75%
- 2-class (stress/non-stress): RF with wrist data ~87%
- Best chest results hit ~93%
Relevance
Closest public dataset to WHOOP signal types. Wrist EDA + BVP → HR, HRV, ACC, TEMP. 300+ citing papers with well-established benchmarks.
Linked To
- Paper: schmidt-2018-wesad
- Physiology: meltdown-signature