An Effective EEG Signal-Based Sleep Staging System using Machine Learning Techniques

Single-channel electroencephalography (EEG) is the most popular choice of sensing modality in sleep staging studies because it widely conforms to sleep staging guidelines. Recently, machine learning methods based on single-channel EEG employ models to capture temporal context among sleep stages. Mainly, the characteristics of brain waves during the sleep stage transition play an essential role in capturing the temporal dependencies across different epochs. To solve these problems, we propose an automated sleep staging system, which refines the sleep stage boundary information based on extracting the multi-scale temporal context between sleep stages. This study has conducted extensive experiments to classify sleep EEG from the PhysioNet dataset in EDF format. 5-class sleep classification has been performed using random forest (RF), K-Nearest Neighbors (KNN), and Random under Sample Boosting (RUSBoost) classifier. The result in the random forest classifier is far better than that obtained by using the KNN classifier. The 5-class classification is also more accurate than most models using a similar database. The training and test dataset is divided into a ratio of 80:20. The five classification stages are the Wake stage, S1, S2, S3, and S4. This classification is based on Rechtschaffen and Kale's (R&K) scoring criteria. The experimental results demonstrate that the RF classifier outperforms state-of-the-art techniques in different evaluation metrics by capturing the changes in characteristics of sleep behavior over the individual sleep stages.