Comparative Analysis of Machine Learning and Deep Learning Algorithms for Automatic Sleep Staging Using EEG Signals

The background and goal of this research are to address the significance of sleep in our lives and the impact of sleep disorders on our health. To analyze sleep problems, legitimate scoring of sleep stages is fundamental, and this is usually finished through a tedious visual survey of, for the time being, polysomnogram by a human expert. However, this process can be improved with artificial intelligence algorithms. To accurately interpret the physiological signals associated with sleep disorders, it is essential to understand how changes in sleep stages are reflected in the signal waveform With this knowledge; automated sleep stage scoring systems can be developed, which not only make the sleep order diagnosis more efficient but also provide insight into the amount of information about sleep stages that can be gleaned from a particular physiological signal. In this study, Single-channel electroencephalography (EEG) is the most commonly used sensory modality in sleep staging investigations, according to this study, because it closely adheres to sleep staging recommendations. A comparison between the machine learning model and the deep learning (CNN) model employing the five-class sleep categorization was undertaken to tackle these difficulties. The machine learning model involves pre-processing, feature extraction, and selection classification based on the stacking model. The deep learning algorithm consists in collecting and pre-processing a labeled dataset of EEG recordings, designing and training a CNN model to classify sleep stages, and testing the model on new data to evaluate its performance. The classification stages are based on Rechtschaffen and Kale (R and K) rules, i.e., Wake, N1, N2, N3, and REM. Based on the obtained results, the machine learning (stacking) algorithm has been found to achieve an accuracy of 97.80%, and the deep learning (CNN) algorithm has been found to achieve an accuracy of 63.30%. The accuracy of the deep learning (CNN) algorithm can be improved using data augmentation techniques, optimizing hyperparameters, and transferring learning, which will be considered in future work.