Title: Predicting sleep quality with digital biomarkers and artificial neural networks

Professor Sungkyu Park presents a novel methodology for predicting sleep quality using digital biomarkers collected from wearable devices and artificial neural networks. Unlike previous studies that primarily relied on short-term HRV measurements in clinical settings, this research developed a model to predict next-day wake after sleep onset (WASO) based on continuous heart rate variability (HRV) data collected from 82 participants wearing Samsung Galaxy Watch Active 2 devices during their daily lives. The study found that among HRV features, the LF/HF ratio showed the strongest correlation with WASO, and the LSTM deep learning model achieved the highest performance with an accuracy of 90.4%. LIME analysis confirmed that the LF/HF ratio, along with Insomnia Severity Index (ISI) and World Health Organization Quality-of-Life Brief Version (WHOQOL-BREF) scores, were the most influential factors for model predictions. These findings provide crucial evidence for establishing public health policies aimed at early detection and prevention of sleep disorders at the population level through personalized sleep health monitoring systems using wearable devices.

Figure: Methodological framework for integrating questionnaire and biometric data in predictive health analytics.

https://www.frontiersin.org/journals/psychiatry/articles/10.3389/fpsyt.2025.1591448/full