A study on sleep EEG Using sample entropy and power spectrum analysis

Abstract

Research on the automation of sleep stage classification, particularly single channel EEG, has been a challenge for many years. The research aims to look into the analysis and evaluation of feature extraction techniques and classification methods that are important to properly classify sleep stages with limited channels. Sample entropy, and the power spectrum of the harmonic parameters using infinite impulse response filters and wavelet transform were used to extract features from the data taken from Physionet database. A total of 13 features were initially extracted and used for the training and testing of the sleep stage classification system. Analysis of the training data showed a distinct combination patterns between the sample entropy and harmonic parameters with a change in the sleep stage. In addition, a prototype for the sleep stage classification system was implemented. Support Vector Machine (SVM) was utilized for the classification system. While the training data were extracted from several database. Further refinement of the data and the program could be useful for a test the sleep stage classification on other database or data.