Extending Stress Detection Reproducibility to Consumer Wearable Sensors

Area of Science:

• Physiological computing
• Biomedical engineering
• Wearable technology

Background:

• Wearable sensors are crucial for collecting physiological data to develop stress detection models.
• Reproducibility of these models across different devices, populations, and study conditions remains a significant challenge.
• Previous work assessed model reproducibility using research-grade devices and physiological signals like heart rate and electrodermal activity.

Purpose of the Study:

• To extend the assessment of stress detection model reproducibility to consumer-grade wearable sensors.
• To compare the performance of research-grade devices (Biopac MP160, Polar H10, Empatica E4) against consumer wearables (Garmin Forerunner 55s).
• To evaluate device-specific stress detection performance in a controlled laboratory setting using undergraduate students.

Main Methods:

• A new stress study was conducted with 35 undergraduate students at Northeastern University.
• Participants completed three standardized stress-induction tasks in a controlled lab environment.
• Physiological data including heart rate variability (HRV) and electrodermal activity (EDA) were collected using various wearable sensors for stress detection model evaluation.

Main Results:

• Biopac MP160 demonstrated the highest performance, consistent with its gold-standard status.
• Combining HRV and EDA generally enhanced stress prediction accuracy across most devices and models.
• Garmin Forerunner 55s showed strong potential for real-world stress monitoring, achieving high accuracy comparable to research-grade devices in specific tasks.
• Empatica E4 exhibited variability, with hardware-model compatibility issues impacting generalizability despite improved performance when combining HRV and EDA in leave-one-subject-out evaluations.

Conclusions:

• Stress detection model performance varies significantly across different wearable devices, including consumer-grade options.
• Combining HRV and EDA data is beneficial for improving stress prediction, but hardware-model compatibility is critical for reliable real-world application.
• Consumer wearables like the Garmin Forerunner 55s offer promising capabilities for stress monitoring in free-living contexts, warranting further exploration for enhanced reliability.