Artificial intelligence for epileptic seizure prediction: Toward hybrid, patient-centric, and edge-ready systems

This review provides a comprehensive examination of artificial intelligence (AI) approaches for epileptic seizure prediction, emphasizing engineering solutions that enhance accuracy, scalability, and clinical applicability. While generalized models, often trained on large multi-patient datasets, have advanced the field, they remain limited in capturing the neurological variability required for reliable patient-specific predictions. This work analyzes recent progress in processing electroencephalography (EEG), electrocardiography (ECG), and electromyography (EMG) signals, and evaluates cutting-edge deep learning methods for feature engineering, classification, and deployment in resource-constrained environments. Key challenges in current systems are identified, including limited availability of patient-specific data, a lack of standardized evaluation practices, and high computational demands that hinder real-time use. The primary AI contribution of this paper is the proposal of a novel hybrid prediction framework. This framework integrates a generalized deep learning backbone for robust feature representation with a lightweight personalization layer, designed to adapt predictions using minimal individual data through techniques like transfer learning and Dynamic Time Warping. The engineering application focuses on enabling this framework for practical, real-world adoption by incorporating feature reduction, model compression, and hardware-aware optimization. This allows for deployment on wearable and edge-computing platforms, ensuring computational efficiency and sustainability. By combining robust generalization with essential personalization, this paper establishes a blueprint for the next generation of sustainable Artificial Intelligence-driven seizure prediction systems and highlights engineering-driven pathways toward clinically viable neurotechnology.