Sleep Stage Classification Using S-Transform-Based Spectral Energy Feature

Abstract

In this paper, we have presented a novel approach to extract the features of EEG signal from its S-transform, in order to classify human sleep disorder accurately. The estimation of sleep disorder plays an important role in identifying sleep stages. Correct predictions can aid doctors in the diagnosis and further treatment. In this paper, features are extracted in both time and time-frequency domain and classified using a various machine learning algorithm. The recorded EEG datasets are divided into delta, theta, alpha, beta, and gamma sub-bands, and a set of descriptive statistical features are attained from all the sub-bands in the time domain. In the time-frequency domain, spectral energy features are extracted from the S-transform. Finally, classification is done using three different classifiers such as ANN, KNN, and SVM. The extensive numerical simulation results illustrate that the proposed method is providing more classification accuracy.