Imbalance Learning Using Neural Networks for Seizure Detection

Around 1% of world's population suffer from epileptic seizures which can lead to injuries and even unexpected death. Making use of EEG signals, which are proven to be the best indicators of seizures, we aim to build an Artificial Neural Networks to classify seizure and non-seizure events. However, the limited availability of seizure events in the EEG data makes it difficult for the automatic classifiers in general to accurately classify seizure events. To improve this, we propose an imbalance learning approach to improve accuracy of highly imbalanced seizure dataset. Since each patient provides a different response to the seizure, we personalize the classification models in terms of training data and model parameters. The proposed imbalance learning method provides an average F-measure accuracy above 86% for Physionet MIT dataset.