Enhanced Sleep Stage Classification Using EEG and EOG: A Novel Approach for Feature Selection with Deep Learning and Gaussian Noise Data Augmentation

Abstract

Accurate identification of sleep stages is critical for understanding its impact on human health. This study introduces a robust method for classifying sleep stages using polysomnography (PSG) data comprising both electroencephalogram (EEG) and electrooculogram (EOG) signals. The initial step in our approach involves extracting signals from the raw PSG data, followed by a preprocessing phase. Following this, feature extraction is executed using wavelet transform, enabling the precise capture of signal attributes such as mean and max values, among others. Utilizing an innovative method known as Autoencoder for Selection (AES) enhances the capability to differentiate between distinct features, thus improving the selection process. In addition, Gaussian Noise Data Augmentation (GNDA) is employed to enhance our dataset and our model’s robustness. Furthermore, various classifiers, including KNN, Bagging, Decision Tree, and FCNN, were used to assess the proposed feature extraction approach. To mitigate overfitting, a 10-fold cross-validation method was utilized. The experimental results indicate that the combination of EEG and EOG with GNDA, AES, and the KNN classifier achieved remarkable performance in sleep stage diagnosis, yielding an accuracy of 97.17%, and an area under the curve (AUC) of 98.2%. Notably, the combined EEG and EOG exhibited superior performance compared to individual models, as well as the existing techniques reported in the literature.