Predicting sleep quality with digital biomarkers and artificial neural networks

Hyolim Lee¹Minsung Cho²Sang Won Lee^3*Sungkyu Shaun Park^4*

• ¹Kangwon National University, Chuncheon, Gangwon, Republic of Korea
• ²Seoul Asan Medical Center, Seoul, Republic of Korea
• ³Kyungpook National University School of Medicine, Daegu, Republic of Korea
• ⁴KDI School of Public Policy and Management, Sejong, Republic of Korea

This study investigates the relationship between heart rate variability (HRV) and sleep quality, emphasizing wake-after-sleep-onset (WASO), a critical marker of sleep fragmentation. By analyzing biometric data, including HRV signals collected from wearable devices and lifestyle data such as daily step counts, alongside subjective questionnaire responses, the study identifies key features associated with WASO. Among these, the low-frequency to high-frequency (LF/HF) ratio emerged as a strong correlate, highlighting its potential as a predictive indicator. Predictive models were developed using HRV, lifestyle, and questionnaire data from the preceding seven days to estimate next-day WASO, integrating behavioral and physiological patterns over a weekly timeframe. A variety of traditional time-series analysis models and machine learning models were applied, with LSTM demonstrating superior predictive performance across key metrics. This work introduces a novel approach for managing sleep health through continuous monitoring and predictive modeling, underscoring the promise of digital biomarkers in delivering personalized, data-driven healthcare interventions for everyday life.