Listen to Your Heart: Stress Prediction Using Consumer Heart Rate Sensors

Galvanic skin response (GSR) is commonly used to measure short-term emotional and cognitive stress, but measuring GSR requires obtrusive equipment. We present a model to predict stress levels solely from electrocardiogram (ECG) data, which can be measured with wearable consumer-grade heart monitors. Our model incorporates time and frequency domain features of heart rate variability, and the spectral power components of the ECG. We make predictions on a sliding window of ECG signal. We apply a linear model to predict stress as a continuous quantity, and our prediction is correlated with the actual GSR with R2 = 0.873 ± 0.035. We also present a model for classifying each window as a binary “stress” or “rest” periods. The best performance was achieved with a linear SVM, with an F1 score of 0.98 on the most distinct samples (highest and lowest 20 percentile GSR levels), and F1 = 0.85 over all samples.