Can Sleep Quality Attributes be Predicted from Physical Activity in Everyday Settings?

Area of Science:

• Physiology
• Biomedical Engineering
• Data Science

Background:

• Sleep is vital for overall health, with established links to physical activity.
• Previous research on physical activity and sleep quality is limited, especially long-term.
• Existing sleep prediction models often focus on single sleep aspects, neglecting comprehensive analysis.

Purpose of the Study:

• To develop a predictive model for sleep duration, efficiency, and deep sleep.
• To investigate the relationship between daily physical activity and sleep quality.
• To identify key physical activity parameters influencing specific sleep metrics.

Main Methods:

• Utilized an Extreme Gradient Boosting (XGBoost) model for sleep prediction.
• Collected data from 34 pregnant mothers using OURA rings over six months under free-living conditions.
• Employed Shapley Additive Explanations (SHAP) to determine feature importance.

Main Results:

• Achieved high prediction accuracies: 90.58% for sleep duration, 95.38% for sleep efficiency, and 91.45% for deep sleep.
• Identified sedentary time as the most influential factor for predicting sleep duration.
• Found that inactive time negatively impacts sleep efficiency, and pregnancy week is crucial for deep sleep prediction.

Conclusions:

• The XGBoost model effectively predicts multiple sleep parameters using wearable-derived physical activity data.
• Physical activity metrics, including sedentary and inactive time, significantly influence sleep quality during pregnancy.
• Pregnancy week is a critical predictor for deep sleep, highlighting its importance in maternal sleep health.