Classification of Intrinsic Mode Spectra of EEG Signal for Epilepsy Detection

In this paper, we proposed EEG based epileptic seizure detection where statistical bandwidth of intrinsic mode functions (IMFs) are used. EEG signals are decomposed into IMFs through the implementation of Empirical Mode Decomposition. These IMFs are turned into analytical functions by means of Hilbert transformation. Euclidean value of Amplitude and Frequency Modulated Bandwidths are calculated from these analytic functions. Finally, the statistical distribution bandwidth of these different IMFs is considered as a feature. The classification was done by means of Support Vector Machine (SVM) with the Gaussian kernel. For training purpose, 75% of the data is utilized. Finally, 89.6% accuracy was gained for the remaining 25% test data in the SVM classifier.