We can’t show the full text here under this license. Use the link below to read it at the source.
Sleep stage prediction with raw acceleration and photoplethysmography heart rate data derived from a consumer wearable device
Predicting sleep stages using movement and heart rate data from a wearable device
AI simplified
Abstract
Our method achieved 90% accuracy in classifying sleep-wake epochs using data from the Apple Watch.
- The best performance for sleep-wake classification was obtained using neural networks.
- Specificity for true wake epochs was 59.6%, while sensitivity for true sleep epochs was 93%.
- Overall accuracy for distinguishing between wake, NREM sleep, and REM sleep was approximately 72% when all features were utilized.
- Models trained on Apple Watch data successfully predicted sleep patterns using external data from the Multi-ethnic Study of Atherosclerosis (MESA).
- This work provides a novel approach to analyzing raw data from wearable devices to enhance sleep prediction accuracy.
AI simplified
Key numbers
90%
Sleep/Wake Classification Accuracy
Percentage of epochs correctly classified as sleep or wake.
93%
True Sleep Epoch Sensitivity
Fraction of true sleep epochs correctly scored.
72%
REM/NREM Classification Accuracy
Accuracy for classifying wake, NREM, and REM sleep.