PCA and SVM Technique for Epileptic Seizure Classification

Abstract

In India we have shortage of skilled Neuro-Physicians who can correctly and timely analyze the complicated features of electroencephalogram (EEG) signal which is critical in epilepsy diagnosis, and hence developing a reliable seizure classification model remains a challenging task. A Support Vector Machine (SVM)-based mechanism is proposed in this paper for classification of epileptic seizures from EEG recordings of brain activity. Certain relevant features are selected from time-frequency domain EEG recordings (TFD). Principal Component Analysis (PCA) technique is applied to improve the performance of the model and for classification SVM classifier with different kernels is applied. According to the results, the proposed PCA-SVM radial basis kernel approach is capable of improving epilepsy classification, as made evident by the results, which show an accuracy of 96.6% for normal subject data versus epileptic data. The performance with other parameters too show promising results hence the proposed SVM-RBF model achieves robust classification for epilepsy.