SleepNet: Attention-Enhanced Robust Sleep Prediction using Dynamic Social Networks.

Area of Science:

• Computational health
• Network science
• Behavioral informatics

Background:

• Sleep behavior is crucial for physical and mental health.
• Ubiquitous sensors can monitor sleep, aiding health management.
• Sleep is influenced by physiology, digital media, social networks, and weather.

Purpose of the Study:

• To develop SleepNet, a system for predicting next-day sleep duration.
• To integrate physiological, phone, and social network data using graph networks.
• To leverage social contagion effects for improved sleep behavior prediction.

Main Methods:

• Proposed SleepNet architecture utilizing graph networks and an attention mechanism.
• Integration of physiological and phone data from mobile and wearable devices.
• Exploitation of social contagion in sleep behavior via network analysis.

Main Results:

• Incorporating social networks significantly enhances sleep duration prediction accuracy.
• SleepNet demonstrates robustness against perturbations in input data.
• Network topology, specifically eigenvalue centrality, impacts prediction stability.

Conclusions:

• SleepNet effectively predicts sleep duration by integrating diverse data sources.
• Social contagion is a significant factor in sleep behavior prediction.
• Understanding network topology is key to robust sleep behavior monitoring systems.