Machine Learning Approaches in Photoplethysmography-Based Sleep Stage Classification

Multi-stage sleep classification is crucial in diagnosing sleep disorders and in evaluating sleep quality, but conventional polysomnography techniques are intrusive and time consuming. This paper examines whether continuous sleep monitoring is feasible using single-sensor-based photoplethysmography (PPG). Several machine learning classifiers including Random Forest, CatBoost, XGBoost, and SVM are used to identify the impact of sleep staging on PPG. These models can perform well in classifying between 4-sleep stages with CatBoost achieving an accuracy of 75.29% similar to the findings of the traditional PSG. The study shows that machine learning approach employed to analyze PPG signals offers a more integrated and accessible solution for the realtime sleep tracking while highlighting the importance of determining the appropriate model for the optimal performance.