Sleep Stage Classification Based on Histogram Gradient Boosting

Sleep stage identification is one of the initial steps in identifying sleep-related disorders. The diagnosis andclassification of sleep stages are performed based on the frequency and nature of the received signals duringpolysomnography (PSG) tests. Electroencephalography (EEG) is a common physiological signal used to monitor brainactivity and diagnose sleep disorders. For this purpose, specialists utilize approximately ۳۰ seconds of EEG signalscalled epochs. Initially, this type of diagnosis is done manually with the assistance of trained experts, which oftenleads to errors and requires significant time and energy. Nowadays, automatic sleep stage detection is carried out usingmachine learning approaches. However, not all neural network methods classify well enough, and due to imbalancedsleep stage data (e.g., S۱), they perform poorly. In this article, we have employed machine learning methods ofBagging, Decision-Tree, and Random-Forest classifiers as state-of-the-art baseline classifiers and compared theirperformances with the proposed Histogram-based Gradient Boosting (HGB) classifier to classify sleep stages into fivedistinct classes. Despite the imbalanced data, we observe that the Histogram-based Gradient Boosting classifiermethod achieves high accuracy in detecting each class of sleep stages for all individuals as compared to other baselinemethods.

Keywords:

sleep stage classification , machine learning , Histogram-based Gradient Boosting , polysomnography , electroencephalography.