Tree-based machine learning models, particularly XGBoost, CatBoost, and Random Forest, provided the strongest overall performance for three-state EEG-based epileptic seizure classification.
Tree-based machine learning models, particularly XGBoost and CatBoost, demonstrate strong performance in classifying epileptic seizure states using EEG sub-band analysis.
Timely identification of seizure-related EEG states can support clinical assessment and motivate future monitoring tools. This study investigates a comparative machine-learning framework for EEG-based epileptic seizure state classification using sub-band analysis of the Bonn University EEG dataset. The signals are decomposed into delta, theta, alpha, beta, and gamma bands with bandpass filtering, and a set of lightweight statistical descriptors is extracted from each band. Support Vector Machines, Random Forest, XGBoost, CatBoost, Multi-Layer Perceptron, k-Nearest Neighbours, and Logistic Regression are then evaluated using multi-class ROC analysis, confusion matrices, precision, recall, F1-score, and accuracy. The results show that tree-based models, particularly XGBoost, CatBoost, and Random Forest, provide the strongest overall performance on this benchmark. The manuscript is now framed as a three-state EEG classification study rather than prospective seizure forecasting, and the discussion highlights the limited size of the Bonn dataset, the need for validation on larger datasets such as CHB-MIT and Siena Scalp EEG, and the requirements for future real-time deployment.
Omran et al. (Wed,) conducted a other in Epileptic seizure. Tree-based machine learning models (XGBoost, CatBoost, Random Forest) vs. Other machine learning models (SVM, MLP, k-NN, Logistic Regression) was evaluated on Three-state EEG classification performance (accuracy, precision, recall, F1-score). Tree-based machine learning models, particularly XGBoost, CatBoost, and Random Forest, provided the strongest overall performance for three-state EEG-based epileptic seizure classification.