EEGNETTINY-SE: A HYBRID MODEL FOR EPILEPSY PREDICTION

Abstract

Epilepsy is one of the most common neurological conditions in which the brain develops abnormal electrical activity, which can cause seizures or episodes. The prompt and accurate diagnosis of seizures is essential for proper treatment planning. EEG is an important modality for capturing real-time brain signals. In this paper, various classifiers, i.e., traditional Machine Learning classifiers and a Deep Learning model based on a Convolutional Neural Network (CNN) and an EEGNetTiny-SE model, are compared for the aim of automatic epilepsy detection from Electroencephalogram (EEG) signals. The data consisted of a synthetic multi-channel EEG recording set with 19 channels, 256 Hz sampling, and 8-second segments, incorporating seizure spike patterns, baseline brain rhythms, and noise components. The statistical and frequency-domain features (delta, theta, alpha, and beta bands; beta power) were extracted for the ML classifiers, and Power Spectral Density (PSD) spectrograms were generated for the CNN classifiers. In addition to applying the proposed EEGNetTiny-SE model, the CNN model, five ML models (Random Forest (RF), Support Vector Machine (SVM), Logistic Regression (LR), Gradient Boosting (GB), and Multi-Layer Perceptron (MLP) were tested. The EEGNetTiny-SE model outperformed the CNN model (85.9% accuracy, 92.28% AUC), as detailed in the comprehensive evaluation based on Accuracy, Precision, Recall, F1-Score, Confusion Matrix, and AUC-ROC, where the EEGNetTiny-SE model achieved 88.7% and 94.1%, respectively. The proposed framework shows promising application potential for intelligent, automated epilepsy diagnostic devices and for real-time monitoring in healthcare applications.