Exploiting Sparse Representation for Sleep Stage Classification Using Electroencephalogram Signal

In this paper, sparse representation of EEG signal is used to automatically classify sleep stages. In this regard, two general sparse representation trends are proposed to classify 4-class sleep stages. The first proposed method is based on sparse principal component analysis (SPCA) which uses different features including time, frequency, and time-frequency features applied to support vector machine (SVM) classifier. The second proposed method is based on sparse representation-based classifier (SRC) which uses orthogonal matching pursuit (OMP) algorithm to obtain sparse coding of the EEG signal. In order to evaluate the effectiveness of the proposed algorithms, their performance is compared with the conventional SVM classification based on PCA method using time, frequency, and time-frequency features. The study is carried out on EEG signal from Physionet international database. Simulation results show on the average 8.36% and 8.26% improvement of the first proposed method in terms of classification accuracy compared to the PCA and deep learning methods, respectively, while the second proposed method has achieved the running time of 118% and 72% faster than the existing PCA and deep learning methods, respectively.