Impact of Multi-domain Features for EEG Based Epileptic Seizures Classification

Accurate detection and classification of epileptic seizures play a pivotal role in clinical diagnosis and treatment. This study introduces an innovative approach that leverages multi-domain features extracted from Electroencephalogram (EEG) data in conjunction with Supervised learning classification techniques. Initially, EEG data undergoes preprocessing through data standardization, followed by the extraction of essential features per instance, encompassing combination of Time domain, Frequency domain, and Time-Frequency domain features. These extracted feature combinations are subsequently fed into the machine learning-based boosting classifier Adaptive Boosting (ADABOOST) for an accurate and precise classification of epileptic signals. Validation of the proposed method is conducted using EEG data from the BEED (Bangalore EEG Epilepsy Dataset) and BONN (University of BONN, Germany) database to detect epileptic seizures. The experimental results show remarkably high levels of classification accuracy for various conditions: 99% accuracy for BEED data, 98% accuracy for BONN data for classifying seizures from healthy states, and 91% accuracy for classifying seizure onset from seizure events. Furthermore, the study applies the Gaussian Naïve Bayes (GNB) classifier to differentiate various types of epileptic seizures, employing evaluation metrics such as the confusion matrix, ROC curve, and diverse performance measures. This method demonstrates significant potential in supporting experienced neurophysiologists decision in the clinical classification of epileptic seizure types.