A Machine Learning Model for Predicting Sleep and Wakefulness Based on Accelerometry, Skin Temperature and Contextual...

Area of Science:

• Biomedical Engineering
• Sleep Science
• Machine Learning Applications

Background:

• Body-worn accelerometers are widely used for estimating sleep duration in large studies.
• Accurate sleep/wake scoring using accelerometry is challenging due to reliance on movement detection.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) model for predicting accelerometry-based sleep duration.
• To assess if adding skin temperature, circadian rhythm, and cyclic time features enhances prediction accuracy.

Main Methods:

• Twenty-nine adults underwent polysomnography (PSG) and dual-accelerometer recordings (AX3, Axivity, UK) with skin temperature sensors.
• PSG sleep/wake scoring served as the ground truth for training the ML model.

Main Results:

• The ML model achieved 0.95 sensitivity and 0.52 specificity using only accelerometer data.
• Incorporating skin temperature and contextual data improved specificity to 0.72, maintaining 0.95 sensitivity.
• Sleep overestimation decreased from 54 min to 19 min with the enhanced model.

Conclusions:

• A dual-accelerometer setup with an ML model can predict sleep/wake periods effectively.
• Adding skin temperature and contextual information significantly enhances the specificity of sleep duration prediction.