Efficient methods for acute stress detection using heart rate variability data from Ambient Assisted Living sensors

Area of Science:

• Biomedical Engineering
• Physiological Computing
• Wearable Technology

Background:

• Ambient Assisted Living (AAL) sensors can detect acute stress, aiding individuals with conditions like diabetes or heart disease.
• Developing reliable stress detection methods is crucial for mitigating the adverse effects of everyday stressors.

Purpose of the Study:

• To develop a reliable stress detection method using heart rate variability (HRV) features from portable electrocardiogram (ECG) sensors.
• To evaluate the impact of different HRV feature sets, time window lengths, and training strategies on stress detection performance.

Main Methods:

• Trained classification algorithms using HRV features extracted from 7 participants' ECG recordings during validated stress tests (Trier Social Stress Test, Stroop test).
• Investigated performance using various HRV feature sets (all, time-domain/non-linear, frequency-domain), time window lengths (overlapping/non-overlapping), and participant-wise training.

Main Results:

• Models utilizing time-domain and non-linear HRV features with 5-min overlapping windows achieved 96.31% accuracy and 96.26% F1 score.
• Participant-wise training yielded high average F1 scores, reaching 99.47% with all features.
• Non-overlapping windows and shorter windows showed reduced performance compared to optimal configurations.

Conclusions:

• Stress detection models using single ECG sensor HRV data perform comparably or better than multi-sensor approaches.
• Reliable stress detection is feasible with portable devices like smartwatches, potentially eliminating the need for additional physiological measurements.