A machine learning application for epileptic seizure detection

Electroencephalography can be treated as an electrophysiological method that can be used to monitor the electrical activity of the brain. Having EEG signal as an aid, there are innumerable diseases that can be detected. Epilepsy is one such disease that can be easily encountered with the abnormalities in the EEG signal. Epilepsy is a condition that affects many people, rendering it the most common neurological disorder after stroke. However it is still difficult to detect some subtle but critical changes in an EEG signal. In this paper we are designing an automated system (classifier) that classifies the recorded EEG signal into Normal, Interictal and Ictal cases. The automation is achieved by extracting various features that include statistical data in the transformed domain using Wavelet and Hilbert techniques. Also the approximate entropy of the sub-bands are included. Now having known the range of the values, each of the features is given a rank. These features are used to enhance the differences between the three cases. Classifier performance is evaluated in terms of its accuracy, specificity and sensitivity. This automated classifier can classify the EEG signal into the desirable cases and has found out its way in biomedical applications by simply repudiating the conventional methods.