Machine Learning-based Feature Extraction Method for Sleep Stage Classification

Abstract

Sleep stage classification is important in diagnosing and treating sleep disorders, but current methods have limitations. An automatic feature extraction method for sleep stage classification using a convolutional neural network (CNN) is proposed. The method extracts channels from raw data, resizes them into sequential functions, and feeds them into a CNN architecture designed to extract relevant features. The proposed method is evaluated on the ISRUC raw sleep dataset, achieving an accuracy of 72.6% on 60 subjects with 11 channels. Comparison with state-of-the-art methods showed that the proposed strategy using PhysioNet Database achieved a higher accuracy of 74%. Learned feature extraction methods are more effective than other preset feature extraction methods. However, challenges still exist, such as using multichannel data and big data. Further research is needed to address these challenges and improve performance. The proposed strategy shows promise in improving the accuracy of sleep stage classification, which can aid in diagnosing and treating sleep disorders.