A robust approach towards epileptic seizure detection

In this paper we present the application of ensemble learning to epileptic seizure detection problem. We propose a robust learning framework to mitigate class imbalance in large CHB-MIT (982 hrs) scalp EEG dataset. The algorithm being used is RUSBoost which is a hybrid data sampling and boosting technique designed especially for skewed classes. The data that is being used in this study has severe class imbalance, with average representation of 0.38% of seizure class to that of 99.62% of non-seizure class. The proposed approach shows the power of RUSBoost in terms of robustness and generalization. We compared our method with the most successful Support Vector Machine (SVM) based approach and report competitive results of 97% seizure detection accuracy, mean detection delay of 2.7s and false detection rate of 0.08 seizure/hr. We also report fast training times of just under three minutes on average for average training data of 21 hrs.