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

Sleep is a crucial aspect to overall health, impacting mental and physical well-being. The classification of sleep stages is an important step to assess sleep quality, and Photoplethysmography (PPG) has been demonstrated to be an effective signal for this task. Recent works in this area usually employ complex methods that may be unfeasible to be deployed in wearable devices. In this work, we present a XGBoost model for sleep-wake classification based on features extracted from PPG signal and activity counts. The performance of our method achieved a Sensitivity of 91.15 ± 1.16%, Specificity of 53.66 ± 1.12%, F1-score of 83.88 ± 0.56%, and Kappa of 48.0 ± 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.