Epileptic Seizure Detection from Imbalanced EEG signal

Despite the development of neurological imaging techniques, the EEG remains the most useful testing tool for the diagnosis of patients with epilepsy. The aim of this paper is to detectthe epileptic seizure from the EEG signal where the feature extraction was realized with the Discrete Wavelet Transform (DWT). The extracted data were rebalanced at first using the Synthetic Minority Over Sampling Technique (SMOTE), than the rebalanced data was given to K-Nearest Neighbors (K-NN)model for the classification. K-NN was used in order to classify abnormal and normal patients. The proposed algorithm is tested on five data sets, each one contains one hundredsingle-channels, four data sets were considered as normal and one as seizure. An average classification accuracy of 99.57 % has been reached and the proposed method surpasses the performances of the state-of-the-art methods using the same EEG benchmark.