Smartwatch Heart Rate Data Predicts Sleep Quality with 90% Accuracy Using AI
Researchers used Samsung smartwatches to track heart rate patterns during the day, then predicted next-night sleep disruptions with 90% accuracy.
Summary
Korean researchers developed an AI system that predicts sleep quality by analyzing heart rate variability from Samsung smartwatches worn during the day. The study tracked 82 participants for up to 28 days, measuring daytime heart patterns to forecast next-night sleep disruptions. Their LSTM neural network achieved 90% accuracy in predicting wake-after-sleep-onset (WASO), a key marker of poor sleep. The LF/HF ratio from heart rate variability emerged as the strongest predictor, with higher ratios indicating more fragmented sleep.
Detailed Summary
Sleep disruption affects up to 50% of adults, but predicting poor sleep nights before they happen could enable proactive interventions. This Korean study demonstrates how everyday smartwatch data can forecast sleep quality with remarkable accuracy.
Researchers tracked 82 participants across winter and summer 2023 experiments, collecting continuous heart rate variability (HRV) data from Samsung Galaxy Watch Active 2 devices worn during waking hours. Participants removed watches at bedtime for charging, so all predictive data came from daytime patterns. The team used seven days of preceding HRV data to predict next-day wake-after-sleep-onset (WASO), the time spent awake after initially falling asleep.
The standout finding was that the LF/HF ratio—a measure of sympathetic versus parasympathetic nervous system balance—strongly correlated with sleep fragmentation. Participants with higher LF/HF ratios experienced nearly double the WASO time (14.9 vs 7.5 minutes). When fed into an LSTM neural network along with sleep questionnaire scores, this achieved 90.4% accuracy in predicting poor sleep nights.
This approach could revolutionize sleep medicine by shifting from reactive treatment to predictive prevention. Instead of addressing insomnia after it occurs, people could receive personalized recommendations based on their daytime stress patterns. The system identified not just who would sleep poorly, but provided explainable insights into why—primarily autonomic nervous system imbalance reflected in heart rate patterns.
The study's strength lies in its real-world applicability using consumer devices, though the relatively young participant pool (average age 25) and short tracking periods limit generalizability to older adults with chronic sleep disorders.
Key Findings
- LSTM neural networks predicted next-day sleep disruptions with 90.4% accuracy using smartwatch data
- LF/HF heart rate ratio was the strongest predictor, with higher ratios doubling wake time
- Seven days of daytime heart patterns provided sufficient data for reliable sleep forecasting
- Consumer Samsung smartwatches generated clinically relevant sleep predictions
- Participants with higher stress ratios had 14.9 vs 7.5 minutes of nighttime wake time
Methodology
Prospective study of 82 participants wearing Samsung Galaxy Watch Active 2 devices during two 26-28 day periods in 2023. Heart rate variability features extracted from daytime PPG signals were used to train multiple machine learning models predicting next-day WASO.
Study Limitations
Study population was predominantly young adults (average age 25), limiting applicability to older populations. Short tracking periods and removal of devices during sleep may miss important nocturnal patterns that influence sleep quality.
Enjoyed this summary?
Get the latest longevity research delivered to your inbox every week.