Generalized filter bank design for sleep stage classification

In this study, binary sleep stage classification (sleep or awake state) was performed using single-channel EEG signal. A new frequency warping function is proposed for this purpose. This function provides a bending function that can proper orientation and depth of the EEG signal frequency content. In this way a generalized filter set of was designed. With the help of this filter set, cepstrum features are extracted. In classification stage, Support Vector Machines (SVM) are employed because of its good performance at binary classification. According to the experimental results, the highest correct classification rate(accuracy) is 98.40%. The result is better than studies which use same database in literature.