Automatic sleep stage classification based on Dreem headband’s signals

In this paper, we propose a system to perform automatic sleep stage classification based on physiological signals acquired by Dreem Headband. These signals contain 4 EEG (FpZ-O1, FpZ-O2, FpZ-F7, F8-F7), 2 Pulse oximeter (Red & Infra-red), and 3 accelerometer channels (X, Y, Z). The dataset used in this study belongs to a challenge competition, namely as Challenge Data and is publicly available on their website. In this work, sleep stages have been scored according to the AASM standard. Features were extracted from the physiological signals after applying a preprocessing step. Each of the EEG and PPG's features is falling into one of the three categories time, frequency, or entropy. Moreover, ancillary features were also extracted from the accelerometer signal. Extracted features were classified by using support vector machine (SVM), K-nearest neighbor and Random forest classifiers. Due to the class imbalance problem, stratified 5-fold cross-validation was performed in order to tune systems parameters. Results show that among the three models as mentioned above, Random Forest has the best performance for the 5-class classification with accuracy: 79.98± 0.70 and kappa 0.7234±0.0095. The proposed model shows promising results, thus the model can be implemented in Dreem headband to differentiate sleep stages efficiently and be used in clinical applications.