Automated Sleep Staging Based on Multi-Physiological Signals Using RF and XGBoost

Sleep quality is closely related to human health, and sleep staging is essential for evaluating and understanding sleep quality. The development of automatic sleep staging technology helps overcome the limitations of manual staging methods. This study aims to develop an efficient automatic sleep staging method using EEG, EOG, and EMG signals. This technique employs machine learning algorithms and analyzes 30-second epochs. Signals are preprocessed with a bandpass filter, and features are extracted from time, frequency, nonlinear, and entropy domains. Random Forest is used for feature selection, while XGBoost is employed as the classifier to train on the selected features. The results indicate that the proposed method attains an accuracy of 88.6%, outperforming other feature selection methods and classifiers, and demonstrating competitive performance compared to existing machine learning approaches. Additionally, an interpretability analysis of the results is conducted.