Deep Learning Algorithm Predicts Sleep Quality With More Than 75% Accuracy: Presented at SLEEP

Researchers have developed a deep learning-based algorithm that can automatically incorporate factors such as daytime activity, heart rate variability, and sunlight exposure to predict a person’s sleep quality with higher accuracy than machine learning or logistic regression analysis, according to a study presented here at SLEEP 2019, the 33rd Annual Meeting of the Associated Professional Sleep Societies (APSS).

"The result that surprised me most in this study is that outside light exposure during the evening showed a more positive correlation with sleep efficiency the following day than other factors," said Kyungmee Park, MD,Yonsei University College of Medicine, Seoul, Republic of Korea. "It was notable that outside light exposure during evening showed a positive effect on sleep efficiency, despite that total light exposure, including inside light, showed a negative effect on sleep efficiency. This implies that inside light exposure, including artificial light exposure, alters sleep of the following day. However, outside light exposure during the evening might aid people to sleep better. We need more speculation about this, however, I think this is a notable result of this study."

The researchers developed an algorithm that can predict sleep quality using deep learning technology. They measured sleep and sleep-related factors including daytime activity, exposure to light, and heart rate variability in 69 healthy participants, then compared the ability of traditional logistic regression analysis, machine learning and their own deep learning algorithm to predict sleep quality based on these factors.

They found that vigorous activity between the time of waking up and noon, exposure to light during any part of the day, and exposure to outside light during any part of the day were significantly correlated with sleep quality in the logistic regression model.

The deep learning algorithm’s accuracy was 87.2%, compared with 79.2% for logistic regression and 83.6% for the machine learning model.

"This study shows that it’s possible to predict sleep status without any manual or subjective data -- only used data gathered automatically from wearable devices -- with more than 75% accuracy," said Dr. Park. "This finding suggests that a fully automatic sleep prediction algorithm can be developed, which can help patients with insomnia monitor themselves in their daily-living without having to perform manual input of data."