Abstract
Background: Epilepsy is one of the brain illnesses known as epileptic seizures which occurs because of the disruption of the electrical communication between neurons. Epilepsy causes abnormalities in electroencephalogram (EEG) signals. Therefore, with observing and declaring the relevant abnormalities in EEG signals, the Epilepsy would be recognized.
Objective: This paper presents a novel classification approach for normal and epileptic electroencephalogram (EEG) signals recorded from healthy and epilepsy persons. Methods: The classification approach is based on time series prediction. Two adaptive neuro-fuzzy networks (ANFIS) are trained to predict one-step-ahead for the EEG time-series data, where one ANFIS is trained on EEG signals of a healthy person and the other on EEG signals of epilepsy. Classification is performed from a window through which all predicted signals were passed. Results: Separability of classes is obtained because of the morphological dissimilarity of the EEG signals in diverse classes, and each ANFIS specializes in the sort of EEG-data on which it is trained with. This approach is performed on thirteen subjects’ EEG signals with classification accuracy rate of about 98%. Conclusion: The classification is performed in the time domain, and there is no need to map the signals to the other prevalent domain such as the frequency domain (like wavelet transform). Besides, the classification does not need any pre-processing on the EEG signals such as feature extraction and dimension reduction.Keywords: Classification, electroencephalography (EEG), epileptic, adaptive neuro-fuzzy network (ANFIS).
Graphical Abstract