| dc.description.abstract | Electroencephalogram (EEG) analysis plays a crucial role in understanding brain activity and diagnosing neurological conditions. Traditional methods often struggle with the complexity and high dimensionality of EEG data. This study addresses these challenges by developing a novel framework that leverages generative self-supervised learning and autoencoder architecture to enhance EEG data analysis. The primary problem lies in the accurate and efficient extraction of meaningful features from EEG signals, which are inherently noisy and complex. The objectives of this research are to improve feature extraction from EEG data using an autoencoder and accurately predict sleep stages using advanced machine learning techniques. The methodology involves preprocessing the EEG data, segmenting it into 30-second epochs, and annotating it according to standard scoring guidelines. An autoencoder is used for feature extraction, followed by the application of Synthetic Minority Oversampling Technique to address the class imbalance. The encoded features are then classified using a robust machine learning model within a TensorFlow environment. Results demonstrate a high average F1-score of 0.97, indicating the effectiveness of the proposed framework. High evaluation metrics, such as Area Under the Curve, Cohen's Kappa Coefficient, and Matthews Correlation Coefficient, further validate the model's performance. This research presents an effective framework for EEG data analysis, combining generative self-supervised learning and autoencoder techniques. Future work will focus on enhancing the autoencoder architecture, and applying transfer learning to diverse datasets. | en_US |
