| dc.description.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. | en_US |
