Identification of seizure from single channel EEG using Support Vector Machine & Hilbert Vibration Decomposition

A well-known neurological brain dysfunction named epilepsy which is caused by recrudescent seizures. Because of higher temporal resolution, brain activities measured by electroencephalography (EEG) are usually utilized for confinement of seizures and distinguishing proof of epileptic dysfunctions. Detection of EEG seizures by using traditional Fourier-based methods and manual interpretation is tedious and challenging because of non-linear and non-stationary dynamics of EEG. In our research, at first, we have done robust statistical analysis to detect and classify the seizure and nonseizure. But, the result was not accurate enough to detect and classify seizure effectively. For identification of ordinary and epileptic EEG measurement, we approached a novel algorithm based on Hilbert vibration decomposition (HVD). HVD accomplishes Hilbert transform demonstration of instantaneous frequency and bring outs mono components that have particular time-differing amplitudes and instantaneous frequencies from non-stationary signals. Least squares support vector machine (LS-SVM) is used for identifying epileptic seizures in this research. In addition, it is attracting for real-time physiological signal processing applications because of its lower mathematical complexity. The classification accuracy of 97.66% was attained on a test, which was conducted on a benchmark EEG data set. In addition, area of 0.9914 under the receiver operating characteristics (ROC) curve utilizing the delta, theta & alpha rhythms.