EEG-Based Seizure Detection and Pre-Ictal Prediction: A Review with Emphasis on Reproducible Modelling

Abstract

Epilepsy is a widespread neurological disorder characterized by recurrent and unpre dictable seizures, creating significant clinical, social, and safety challenges for affected individuals. Electroencephalography is the primary clinical tool for observing seizure activity, and recent research has increasingly applied machine learning techniques to support automatic seizure detection and short-term pre-ictal seizure prediction. However, reported results vary widely across studies due to differences in datasets, signal processing pipelines, modeling choices, and evaluation protocols, raising concerns about reproducibility and real-world applicability. The objective of this study is to systematically review electroencephalography-based seizure detection and pre-ictal prediction methods, with particular emphasis on modeling practices that influence reproducibility and deployment readiness. A structured literature review methodology was used to analyze public datasets, preprocessing approaches, feature representations, and machine learning and deep learning models, while maintaining a clear distinction between seizure detection and seizure prediction tasks. Key design parameters, including pre-ictal window duration, prediction of horizon, and evaluation of metrics, were examined to assess their impact on reported performance. The review finds that seizure detection methods generally achieve strong performance under controlled conditions, whereas pre-ictal prediction results remain less stable and highly sensitive to experimental design choices. In addition, incomplete reporting of pipelines, limited cross-dataset evaluation, and minimal consideration of resource constraints hinder reproducibility and deployment on mobile or wearable systems. The study concludes that improved reporting standards, structured metadata, and integration of lifecycle management and edge computing principles can significantly enhance transparency, comparability and practical usability of electroencephalography-based seizure analysis systems, contributing to more reliable clinical decision support and continuous monitoring solutions.