A machine-learning sleep-wake classification model using a reduced number of features derived from photoplethysmography and activity signals

TL;DR

This study tackles the problem of sleep–wake classification using wearable signals by employing a XGBoost-based model that relies on a minimal three-feature set derived from photoplethysmography (heart rate and heart rate variability) and activity (actigraphy). Using the MESA Sleep dataset, the authors construct 30-second windows and evaluate three classifiers, achieving a maximum accuracy of 77.6% with strong sensitivity (≈91%) but moderate specificity (≈54%), and a Cohen’s kappa of ~0.48, indicating substantial agreement with sleep-wake labels. The key contribution is showing that a compact feature set can yield competitive performance while enabling deployment on computation-constrained wearables, potentially enabling scalable, at-home sleep monitoring. The work also highlights limitations such as class imbalance and the need for external validation, and suggests future directions in handling imbalance and assessing resource usage for real-device deployment.

Abstract

Sleep is a crucial aspect of our overall health and well-being. It plays a vital role in regulating our mental and physical health, impacting our mood, memory, and cognitive function to our physical resilience and immune system. The classification of sleep stages is a mandatory step to assess sleep quality, providing the metrics to estimate the quality of sleep and how well our body is functioning during this essential period of rest. Photoplethysmography (PPG) has been demonstrated to be an effective signal for sleep stage inference, meaning it can be used on its own or in a combination with others signals to determine sleep stage. This information is valuable in identifying potential sleep issues and developing strategies to improve sleep quality and overall health. In this work, we present a machine learning sleep-wake classification model based on the eXtreme Gradient Boosting (XGBoost) algorithm and features extracted from PPG signal and activity counts. The performance of our method was comparable to current state-of-the-art methods with a Sensitivity of 91.15 $\pm$ 1.16%, Specificity of 53.66 $\pm$ 1.12%, F1-score of 83.88 $\pm$ 0.56%, and Kappa of 48.0 $\pm$ 0.86%. Our method offers a significant improvement over other approaches as it uses a reduced number of features, making it suitable for implementation in wearable devices that have limited computational power.

Figures (3)

• Figure 1: Overview of the proposed method for sleep-wake classification.
• Figure 2: Windowing and Feature Extraction process example.
• Figure 3: Stratified F1-score results for sleep-wake classification task for age and gender.