EEG-Based Epilepsy Prediction: Evaluation Metrics, Data Deficiency and Limitation of Current Methods

Epilepsy is a disease of the brain that can do severe damage for patient's health. Due to the recurrent and unpredictable characteristics of epilepsy, it is of vital importance to develop reliable methods to predict seizures in advance. Nowadays, many researchers have developed deep learning (DL) or machine learning (ML) methods to predict epileptic seizures with electroencephalogram (EEG). But there are still many problems and challenges on the way towards a high-performance and generalized model. This study discussed and analyzed the current ML and DL techniques used in seizure prediction and summarized some challenges that remains to be solved, including the Inconsistency of the evaluation metrics, the imbalance and insufficiency of the available data and some limitations of current models. This study summarized the solutions that used to solve them and proposed some suggestions that can help improve the performance of the models. Furthermore, this review discussed some potential DL/ML methods that can be applied in the area of seizure prediction. This study aims to provide researchers with clear concepts in future works and proposed future directions.