Issues and Causal Factors in Machine Learning Methods for Predicting Epileptic Seizures Using EEG: Towards Clinically Viable Prediction via Utility–Latency Feasibility Constraints

Abstract Approximately one in three people with epilepsy have drug-refractory epilepsy, which means they continue to have seizures despite medication. This neurological disorder, characterized by abnormal electrical activity in the brain, can lead to involuntary movements, loss of consciousness, and even death. In this review, we identify the challenges involved in developing machine learning models for EEG-based seizure prediction, with emphasis on critical stages including dataset design, preprocessing, feature extraction, classification, and post-processing. The main contribution of this work is to propose a classification of these problems based on the development process, distinguishing between pipeline-based and non-pipeline-based problems, with the aim of guiding current research toward real-world implementation. We explore the underlying causes of these problems and highlight proposed solutions from the literature. In conclusion, there is a pressing need to develop algorithms that achieve an optimal balance between efficiency and performance while ensuring interpretability. Additionally, establishing a standardized framework for their comparison is essential. Keywords: issues, seizure prediction, machine learning, EEG, epilepsy Poster accepted: Medina, J. & Serrano, S. Efficiency and Performance in Machine Learning Methods for Epileptic Seizure Prediction. Poster presentation, Institute of Epilepsy Research Conference, UK, 2025. Ranked among the top 7 posters globally. Poster accepted: Medina, J. & Serrano, S. Efficiency and Performance in Machine Learning Methods for Epileptic Seizure Prediction: Issues and Causal Factors. Poster presentation, 36th International Epilepsy Congress (CEC-ILAE), Portugal, 2025.