ANALYSIS OF PROBABILISTIC NEURAL NETWORKS WITH DIMENSIONALITY REDUCTION FOR EPILEPSY CLASSIFICATION FROM EEG

Epilepsy is widely regarded as one of the most common brain disorders in the human beings. It is often accompanied by various disturbances in the behaviour, cognitive impairment and the brain dysfunction. As it is a chronic neurological disease, the hallmark of this disease is recurring seizures. For diagnosis of epilepsy, Electroencephalography (EEG) signals are widely used. The representation of voltage fluctuations which are due to the flow of neurons ionic current is expressed as an EEG signal. As the recordings of the EEG signal are quite long, the dimensionality of it is reduced with the help of Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA) and Singular Value Decomposition (SVD). The dimensionally reduced values are then fed inside the Probabilistic Neural Network (PNN) for epilepsy classification from EEG signals. Results show that when PCA is utilized with PNN an average classification accuracy of 97.01% is obtained, when LDA is utilized with PNN an average classification accuracy of 97.42% is obtained and when SVD is utilized with PNN an average classification accuracy of 92.67% is obtained.