Abstract

Automated seizure detection in electroencephalography (EEG) recordings is a time consuming task, dependent on the expert performing the review. The rapid development in the field of deep learning shows promise in the creation of automated models performing EEG signal classification. However, EEG data requires careful artifact removal, as well as strategies for dealing with inherent imbalance present within samples extracted from epileptic patients. In this work, a simple graph neural network (GNN) using attention to perform classification of EEG segments is proposed. We also elaborate on the effectiveness of signal pre-processing and imbalance handling methods, showing their impact on the model's performance. The results demonstrate that the classificator's performance can be enhanced by choosing proper pre-processing and signal balancing methods. We anticipate that these approaches can be adopted by the researchers working on EEG classification with deep learning models, helping to improve the the robustness of constructed models.