Seizure Classification on Epileptic EEG Using IMF-Entropy and Support Vector Machine

Various methods have been developed by researchers to recognize brain abnormalities through EEG signals. One of the diseases or disorders of the brain is seizures in epilepsy. EEG signals in seizure conditions display a different pattern compared to EEG signals in normal conditions. Researchers analyzed the EEG signal using a variety of observed approaches. One phenomenon used to analyze EEG signals is signal complexity. Signal complexity captures fluctuating patterns of EEG signals quantizing them to distinguish normal and seizure signal conditions. In this study, we propose the proper feature extraction method based on the basic characteristic of the signal. We extract the EEG signal’s information using entropy calculation from the intrinsic mode function (IMF entropy). Our main goal is to distinguish normal and seizure EEG signals. The entropy is calculated from the IMF resulted from empirical mode decomposition (EMD), then entropy from the relative energy of each IMF. To test the performance of the proposed feature extraction method, the support vector machine (SVM) is used as a classifier. The highest accuracy is 86.3%, sensitivity is 86.33%, and the specificity is 93.17% for three data classes: normal, interictal, and seizure. The proposed method has the potential to improve its performance, considering there are still many variations of EMD methods and decomposition levels that can be evaluated. Furthermore, testing on more massive datasets is interesting to do in future research.