Best wearable datasets for stress and emotion detection
WESAD is the reference dataset for wearable stress detection: 15 subjects in a lab protocol with baseline, stress and amusement conditions, recorded with an Empatica E4 on the wrist and a RespiBAN on the chest. Its authors report up to 93% accuracy for stress versus non-stress. UBFC-Phys (56 participants, a TSST-inspired protocol with wrist BVP, EDA and face video) is larger, and K-EmoCon (32 participants in naturalistic debates, with self, partner and observer emotion ratings) covers emotion in social interaction.
Top picks and when to use each
Wrist BVP 64 Hz, EDA and temperature 4 Hz, accelerometer 32 Hz; chest ECG, EDA, EMG, respiration, temperature and acceleration at 700 Hz; plus questionnaires. About 2.5 GB.
Access: Open (scientific, non-commercial use with credit)Original host56 participants, rest, speech and arithmetic tasks, wrist BVP and EDA plus video, with anxiety scores before and after.
Access: IEEE DataPort / dat@UBFCOriginal host16 paired debates of about 10 minutes with E4, Polar H7, EEG and audiovisual recordings; arousal-valence and 18 categorical emotions rated every 5 s.
Access: Restricted (request via Zenodo)Original host- #4CLAS
CLAS (Cognitive Load, Affect and Stress) supports automated recognition of emotion, stress, and cognitive load. Authors: Markova / Ganchev / Kalinkov (TU Varna). Paper: BIA-2019 10.1109/BIA48344.2019.8967457.
Multi-modal continuous stress detection of nurses in a real hospital during COVID-19. Dryad DOI 10.5061/dryad.5hqbzkh6f. ~1,250 hours across two 2020 sessions.
Comparison table
27 datasets. Swipe sideways on mobile. Values come from each dataset record; open a dataset for sources. Popularity is cached at build time: OpenAlex citations of the source paper (or the dataset DOI when no paper is linked), results on ChatPPG leaderboards, and GitHub stars or Zenodo / Hugging Face downloads where the host reports them. Hover for the source; a dash means unknown.
| Signals | Sampling | Ground truth | Licence | Access | |||
|---|---|---|---|---|---|---|---|
| WESAD | 15 | PPG, ECG, accel, skin temp, EDA | 64Hz, 32Hz, 4Hz | stress, mwl | Scientific, non-commercial use with credit to the … | Open (scientific, non-commercial use with credit) | 1,337 cited12 results |
| UBFC-Phys | 56 | EDA, PPG bvp, video face | — | stress, mwl | See host | IEEE DataPort / dat@UBFC | 174 cited |
| K-EmoCon | 32 | PPG, accel, EDA, skin temp, ECG | 64 Hz, 32 Hz, 4 Hz | stress | Restricted access (request on Zenodo); no license … | Restricted (request via Zenodo) | 173 cited4,468 downloads |
| CLAS | 62 | PPG, ECG, accel, EDA | — | stress, mwl | See host | See host | 1 cited |
| NURSE (hospital stress, E4, Dryad) | 15 | PPG, accel, skin temp, EDA | 64Hz, 32Hz, 4Hz | stress, mwl | See host | See host | 2 cited |
| A Multimodal Dataset for Investigating Working Memory in Presence of Music | — | PPG, ECG, EDA, EEG, resp | — | mwl | See host | See host | 0 cited |
| ADARP (AUD EMA stress, E4) | 11 | PPG, accel, skin temp, EDA | 64Hz, 32Hz, 4Hz | stress, mwl | CC BY 4.0 | See host | 13 cited1,354 downloads |
| AQUASET | 30 | PPG, ECG, accel, skin temp, EDA | 64 Hz | stress | See host | See host | 0 cited5 downloads |
| BIOSTRESS (psychologist induction + exam) | 20 | PPG, accel, skin temp, EDA | 64Hz, 32Hz, 4Hz | stress, mwl | See host | See host | 0 cited |
| CAN-STRESS (ambulatory E4 stress 0–9) | 82 | PPG, accel, skin temp, EDA | 64Hz, 32Hz, 4Hz | stress, mwl | See host | See host | 1 cited224 downloads |
| E4 Stress + Exercise (PhysioNet wearable-device-dataset) | 36 | PPG, accel, skin temp, EDA | 64Hz, 32Hz, 4Hz | stress, mwl | See host | See host | 5 cited |
| EmoWear (E4 SAM affect) | 48 | PPG, accel, skin temp, EDA | 64Hz, 32Hz, 4Hz | stress | See host | See host | 2 cited3,435 downloads |
| EmpaticaE4Stress (workplace cognitive stress) | 29 | PPG, accel, skin temp, EDA | 64Hz, 32Hz, 4Hz | stress, mwl | See host | See host | 1 cited |
| EPIStress (epilepsy cognitive stress, E4+Muse) | 20 | PPG, accel, skin temp, EDA, EEG | 64Hz, 32Hz, 4Hz | stress, mwl | CC BY 4.0 | See host | 1 cited0 stars230 downloads |
| ESCOLAR | 26 | PPG, accel, gyro, skin temp, EDA | 25 Hz | stress, mwl | See host | See host | 0 cited8 downloads |
| Galaxy Watch Active2 raw PPG+ACC (PHQ-9 / GAD-7) | — | PPG, accel | — | mwl | See host | See host | — |
| GroupAffect-4 | 40 | PPG, accel, skin temp, EDA, audio | 25 Hz, 50 Hz, 48 kHz | stress | See host | See host | 0 cited147 downloads |
| MEFAR | 23 | PPG, accel, skin temp, EDA, EEG | 64Hz, 32Hz, 4Hz | mwl | See host | See host | 0 cited |
| Multimodal Micromobility Stress Dataset | — | PPG, EDA, skin temp, accel | — | stress | CC BY 4.0 | See host | 0 cited |
| Office Everyday MWL (MW_OFFICE, n=10 E4) | 10 | PPG, accel, skin temp, EDA, EEG | E4 standard BVP/ACC/EDA/TEMP/HR/IBI; Mus… | stress, mwl | CC BY 4.0 | See host | 0 cited2 stars118 downloads |
| Podcast playback-speed physio | 62 | PPG, EDA, EEG | 8-ch EEG + EDA + PPG/BVP (rates in depos… | mwl | See host | See host | 0 cited11 downloads |
| Regulation of Brain Cognitive States through Auditory, Gustatory, and Olfactory Stimulation with Wearable Monitoring | — | PPG, accel, EDA, EEG | — | mwl | See host | See host | 0 cited |
| SONATA | 50 | PPG, ECG, skin temp, EDA, resp | 256 Hz, 250 Hz | stress | See host | See host | 0 cited26 downloads |
| Stress-Predict Dataset (E4, MIT) | 35 | PPG, accel, skin temp, EDA | 64Hz, 32Hz, 4Hz | stress, mwl | See host | See host | 104 cited16 stars |
| Tutunji prolonged stress | 83 | PPG, accel, EDA, skin temp, HR | — | stress | See host | See host | 34 cited |
| UNIVERSE (mental workload + stress, ~315h E4) | 24 | PPG, accel, skin temp, EDA, EEG | 64Hz, 32Hz, 4Hz | stress, mwl | CC BY 4.0 | See host | 3 cited1,454 downloads |
| MOODS | 122 | PPG, accel | 100 Hz, 50 Hz | stress, ema, activity | See host | See host | 27 cited |
Benchmark results
Reported results from papers in the ChatPPG hub. Protocols differ, so only compare rows on the same dataset and metric.
| Dataset | Metric | Method | Result | Protocol |
|---|---|---|---|---|
| WESAD | Mean absolute HR error (BPM) ↓ | ECG-PPG JEPA 345M | 4.58 | LOSO, frozen encoder (KDD'26 Table 2) |
| WESAD | Mean absolute HR error (BPM) ↓ | GPT-PPG | 4.98 | GPT-1B fine-tuned (MOFT), LOSO; arXiv v1 Table 2 |
| WESAD | Accuracy ↑ | Cross-Modal Correlation Fusion for End-to-End EC | 0.9034 | WESAD subject-independent 5-fold CV; mean Accuracy 90.34%; Stress Recall 94.07% |
| WESAD | Accuracy ↑ | TF-MambaNet | 0.81 | Binary stress, 8 s, frozen features (paper Table 2); accuracy — not comparable to AUROC rows; PaPaGei 0.80 / Chronos 0.7… |
| WESAD | Area under ROC ↑ | Cross-Modal Correlation Fusion for End-to-End EC | 0.9729 | WESAD subject-independent 5-fold CV; dual ECG+PPG encoders + cross-modal correlation fusion; paper AUC-ROC 97.29% |
| WESAD | Area under ROC ↑ | Pulse-PPG | 0.9687 | Binary stress; paper non-stress definition |
| WESAD | Area under ROC ↑ | Thapa wrist-PPG features | 0.965 | LOSO pooled AUC, BVP 64 Hz 60 s windows, stress from 9 subjects; post hoc; interval-only 0.911–0.933; RF 0.972 transduct… |
| WESAD | Area under ROC ↑ | TEX-MEX 0.56M | 0.9630 | Binary stress, subject-level 5-fold CV × 15 seeds (arXiv 2610.19557 Table 4); same-paper Pulse-PPG 0.9388 / AnyPPG 0.952… |
| WESAD | Area under ROC ↑ | Quality-aware PPG+ACC | 0.956 | LOSO 15 subj, 60-s windows, baseline vs stress; low-tier venue; selective prediction improves accepted-window F1 |
| WESAD | exact_match ↑ | PulseLM | 0.497 | 10 s @125 Hz, subject-disjoint ~8/1/1 per source; categorical QA exact-match (arXiv 2603.03331 Table 2); stress label se… |
Preprocessing pitfalls
- WESAD has 15 subjects. Use leave-one-subject-out; window-level splits give near-perfect but meaningless scores.
- Lab stress is not daily-life stress. Models trained on lab stress protocols often fail in field data where stress is self-reported and confounded by activity.
- Many papers drop the amusement class and report binary stress; state 2-class or 3-class.
- WESAD and PPG-DaLiA use the same wrist device and lab group; they are not independent cohorts for generalisation claims.
FAQ
What is WESAD?
WESAD (Wearable Stress and Affect Detection) is a public multimodal dataset of 15 subjects recorded with wrist (Empatica E4) and chest (RespiBAN) devices during baseline, stress and amusement conditions, introduced at ICMI 2018 by Schmidt et al.
What accuracy is reported on WESAD?
The original paper reports up to 80% accuracy for three classes (baseline, stress, amusement) and up to 93% for stress versus non-stress. Newer results are listed in the leaderboard below.
Can WESAD be used commercially?
The dataset page says it may be used for scientific, non-commercial purposes with credit to the owners.