An effective approach to classify epileptic EEG signal using local neighbor gradient pattern transformation methods

Electroencephalographic (EEG) signal records the neuronal activity in the brain and it is used in the diagnosis of epileptic seizure activities. Human inspection of non-stationary EEG signal for diagnosing epilepsy is cumbersome, time-consuming and inaccurate. In this paper an effective automatic approach to detect epilepsy using two feature extraction techniques namely local neighbor gradient pattern (LNGP) and symmetrically weighted local neighbor gradient pattern (SWLNGP) are proposed. Extracted features are fed into machine learning algorithms like k-nearest neighbor (k-NN), quadratic linear discriminant analysis, support vector machine, ensemble classifier and artificial neural network (ANN) to classify the EEG signals. In this study, the classification performance for 17 different cases using 10-fold cross validation with the following classification problems are executed (i) healthy-ictal, (ii) interictal-ictal, (iii) healthy-interictal, (iv) seizure free-ictal and (v) healthy-interictal-ictal. The experimental result shows that in all the cases LNGP and SWLNGP attained higher classification accuracy using ANN. Further, the computational performance and the classification accuracy of the proposed methods are compared with the recently proposed techniques for epileptic detection. It shows that the performance of LNGP and SWLNGP method with ANN classifier are superior over other recently proposed techniques for the aforesaid problems. Hence, the proposed methods are simple, fast, reliable and easily implementable for real-time epileptic detection.