AI-Driven Real-Time EEG Analysis for Epileptic Seizure Prediction using Optimal Features

Abstract

Over 50 million people worldwide suffer from epilepsy, making it one of the most prevalent neurological disorders. Accurate identification and prediction of epileptic seizures are crucial for reducing health-related risks, improving patient safety, and enabling timely clinical intervention. Electroencephalography (EEG), a non-invasive technique for monitoring the brain's electrical activity, has become the primary modality for automated seizure detection and prediction. Conventional machine learning and deep learning approaches have demonstrated promising performance; however, EEG signals are inherently high-dimensional, noisy, nonlinear, and non-stationary, making accurate seizure prediction a challenging task. Consequently, recent research has shifted towards the integration of advanced Artificial Intelligence (AI) techniques, including deep learning, attention mechanisms, optimization algorithms, reinforcement learning, explainable AI, and hybrid intelligent framework. Which achieve moderate accuracy which can be improved. These models fail to learn complex spatiotemporal dependencies; also fail to address the problem of feature redundancy. To address these problems, this work proposes a real-time AI-powered analysis of EEG signals to optimize epileptic seizure prediction. The model incorporates a spatio-temporal attention (STAtt) mechanism for extracting time and space correlation patterns, an enhanced multi-objective bean optimization (EMBO) algorithm, and hierarchical dual-Q learning (HDQL) model for effective seizure detection. The proposed approach would be verified experimentally on the three benchmark EEG datasets: CHB-MIT, Bonn, and TUH. The findings reveal that the incorporation of STAtt, EMBO, and HDQL remarkably contributes to EEG feature discrimination, minimizes misclassification, and offers real-time seizure-detection solution, showing superiority over state-of-the-art models across multiple datasets.