Sleep Stage Classification From Polysomnography (PSG) Data

Abstract

Sleep staging is an essential aspect of sleep assessment and disease diagnosis. Polysomnography(PSG) is commonly used to record electroencephalography(EEG) signals, which produce distinct patterns during different sleep stages. Automatic identification of these stages is crucial in diagnosing and treating sleep related disorders. This study focuses on the classification of sleep stages using polysomnography (PSG) data. The main objectives are to predict five-class sleep stages (Wake, N1, N2, N3, and REM) using EEG data from the Haaglanden Medisch Centrum (HMC) dataset, comparison Of light, deep, and wake sleep stages, and examine the predictive power of subject data from the Sleep Physionet dataset in predicting the sleep stages of other subjects. The dataset contains EEG sleep recordings of 154 subjects of which only 50 were investigated by using two machine learning models which are: Random Forest Classifier and XG Boost. The results and findings of this research indicate that the Random Forest Classifier algorithm achieved an accuracy of 82.62%, outperforming the XG Boost algorithm in sleep stage classification using the given dataset. The findings suggest that the Random Forest Classifier is a more effective model for sleep stage classification using PSG data. The study demonstrates the potential of using machine learning techniques for accurate sleep stage prediction. Further research can explore the generalizability of the model across different datasets and investigate the impact of additional features on classification accuracy. Our research makes contribution by investigating the application of Event-Related Potentials (ERP) in effectively categorizing sleep stages based on PSG data. ERP, which measures neural activity in response to specific stimuli, offers valuable insights into the underlying cognitive processes during sleep. By integrating ERP analysis into our classification approach, we aimed at enhancement of precision and interpretability of sleep stage identification. Our findings highlight the significance of incorporating ERP analysis as a supplementary technique alongside traditional EEG features, thereby advancing the understanding and diagnosis of sleep-related disorders.