Abstract

Epilepsy is a chronic neurological disorder characterized by recurrent seizures resulting from abnormal neuronal activity. Accurate identification of the pre-ictal phase, which precedes seizure onset and contains critical neurophysiological biomarkers, is essential for seizure prediction and early intervention. This study introduces an artificial intelligence (AI)-driven decision support system, termed Epileptic Seizure Scan (ESS), for automated pre-ictal EEG pattern detection. The ESS framework integrates signal preprocessing, wavelet-based time-frequency analysis, entropy-driven feature optimization, dimensionality reduction, and deep learning modules within a unified software environment. For classification, Bidirectional Long Short-Term Memory (BiLSTM) networks were employed due to their ability to effectively model temporal dependencies in EEG signals. To enhance model robustness and mitigate over-complexity, network configurations were optimized using Information Complexity Criteria (ICOMP) in conjunction with conventional model selection approaches. The proposed system was validated on the CHB-MIT Scalp EEG database, demonstrating high performance in pre-ictal pattern recognition with strong accuracy, sensitivity, and precision. These findings underscore the potential of ESS as a reliable AI-assisted clinical decision support tool for seizure forecasting and epilepsy management.