Performance analysis of classifiers in the detection of seizer and normal stages from EEG signals

Abstract

The Electroencephalogram (EEG) signal is popularly used technique to analyse the brain activities and measured using Scalp electrodes which is quite useful in detecting status of the brain and epilepsy of a subject and also supplement the CT scan measurement. EEG signals are indirectly indicate the condition of Brain. EEG signals are sampled at 256 samples/second for a sixteen epochs are considered for this work. In this paper the features of normal and seizure EEG signals are extracted using statistical technique and classified using Detrend Fluctuation Analysis, Detrend Fluctuation Analysis EM, Detrend Fluctuation Analysis log regression, Detrend Fluctuation Analysis Non Linear regression, Detrend Fluctuation Analysis Linear regression. The classifier performance is compared using the standard parameters like sensitivity, specificity, Accuracy, Performance Index and False Alarm. The highest accuracy of 98.96% obtained for the hybrid classifier Detrend Fluctuation Analysis Linear regression for Normal EEG signal and an accuracy of 98.96 % achieved using Detrend Fluctuation Analysis with Expectation Maximization classifier for seizure EEG signal.