Improved Deep Learning Model for Epileptic Seizure Detection Using CNN-BiLSTM

Epileptic seizures are acute, sudden neurological occurrences that can potentially affect a patient’s health and wellbeing. Timely intervention is required, and it can be achieved by early and precise detection from electroencephalogram (EEG) signals. Deep learning models, particularly recurrent and convolutional neural networks, have proven to be highly efficient in processing complex biomedical signals, with a higher feature extraction capability compared to traditional techniques. Ensemble models using BiLSTM and machine learning classifiers like Support Vector Machines (SVM), Random Forest (RF), and XGBoost (XGB) have been proposed in previous paper. The models were highly accurate but used a hybrid model that utilized a combination of deep learning and machine learning classifiers. This paper introduces an optimized ensemble architecture for deep learning that combines Convolutional Neural Networks (CNNs), Bidirectional Long Short-Term Memory (BiLSTM), and adaptive attention mechanism to improve seizure diagnosis. The proposed model employs deep feature extraction and adaptive weighting to enhance the efficiency of classification. The model, which is trained on the Bonn University EEG dataset, achieves 98.75% validation accuracy, 100% recall, 95.23% precision, and 97.56% F1-score, outperforming traditional ensemble methods. These results indicate that the proposed method offers a highly reliable, scalable, and clinically deployable solution for epileptic seizure detection.