malansi/gsr-hrv-stress-detection-dataset

GSR + HRV Stress Detection — Cleaned Volunteer Dataset

Cleaned, anonymized physiological recordings from 124 volunteers used to train the GSR+HRV stress detection model. Collected as part of an applied research project at Almutlaq United Company.

Full data-cleaning pipeline and training code: github.com/Alansi775/GSR-COSSINUS_VOLUNTEERS_DATA_FOR_ML

Protocol

Each volunteer completed an identical 4-phase session, recorded at 1 Hz:

Stage	Duration	Purpose
Calibration	≈30s	Sensor settle / per-volunteer baseline
Normal	≈4min	Relaxed baseline
Stress	≈3min	Scripted time-pressure task
Relaxation	≈1min	Recovery

Files

One CSV per volunteer: cleaned_data_<id>.csv, plus a diagnostic plot pair per volunteer (plot_<id>.png, plot_<id>_nn_interval.png).

Column	Description
Time_sec	Seconds since session start
Conductance_microS	Galvanic skin conductance (µS)
Stage	Protocol stage label (Calibration, Normal, Stress, Relaxation)
RR_Interval_ms	Beat-to-beat heart interval (ms), sparse
Conductance_microS_Normalized	Per-volunteer baseline-normalized conductance
RR_Interval_ms_Normalized	Per-volunteer baseline-normalized RR interval

No names or personal identifiers are included — volunteers are referenced only by numeric/alphanumeric ID.

Known Data-Quality Notes

• Volunteers 35–144: sensor originally logged electrical resistance, corrected to conductance via conductance = 1 / resistance.
• 14 volunteers have RR_Interval_ms populated for fewer than 10 rows — excluded from model training (see the model card for the full exclusion list), but left in this dataset for transparency.

License

cc-by-nc-4.0 — non-commercial use, with attribution.

Author

Mohammed Saleh Almutlaq United Company