Enhancing Sleep Stage Classification With Single Channel EEG: Feature Extraction and Random Forest-XGBoost Model

Abstract

Sleep stage classification, essential for diagnosing sleep disorders, monitoring neurological health, and advancing cognitive neuroscience. Traditionally involves manual scoring of polysomnographic (PSG) recordings, which is time-consuming and requires expert knowledge. However, recent machine learning advancements have enabled automated sleep stage classification using single channel electroencephalographic (EEG) signals. This study focuses on classifying sleep stages with single Channel EEG using an ensemble model combining Random Forest (RF) and XGBoost, achieving high accuracy across five sleep stages: Wake (W), N1, N2, N3, and REM (R). The RFXGBoost model demonstrated robust performance with accuracies of 97.3% for W, 98.6% for N2, 97.5% for N3, and 97.5% for R. However, the model showed lower accuracy for stage N1, achieving a recall of 83.5% and a precision of 57.4%, reflecting challenges in detecting this transitional stage. Receiver Operating Characteristic (ROC) analysis revealed excellent classification capability, with AUC values of 1.00 for W, N2, and N3, 0.99 for R, and 0.97 for N1. Compared to other models, including LSTM, Neural Network (NN), and Support Vector Machine (SVM), the RFXGBoost ensemble achieved the highest overall accuracy of 96.11% and a Cohen's Kappa of 0.9388. By implementing the RFXGBoost model, this study contributes to Sustainable Development Goal (SDG) 3 by supports early diagnosis and monitoring of sleep-related disorders, promoting better health outcomes and contributing to improved well-being. Additionally, it also supports SDG 9. Through data-driven insights, this research contributes to the development of intelligent health monitoring systems, ultimately supporting sustainable healthcare practices.