Classification of EEG Signals Using Sampling Techniques and Least Square Support Vector Machines

This paper presents sampling techniques (ST) concept for feature extraction from electroencephalogram (EEG) signals. It describes the application of least square support vector machine (LS-SVM) that executes the classification of EEG signals from two classes, namely normal persons with eye open and epileptic patients during epileptic seizure activity. Decision-making has been carried out in two stages. In the first stage, ST has been used to extract the representative features of EEG time series data and to reduce the dimensionality of that data, and in the second stage, LS-SVM has been applied on the extracted feature vectors to classify EEG signals between normal persons and epileptic patients. In this study, the performance of the LS-SVM is demonstrated in terms of training and testing performance separately and then a comparison is made between them. The experimental results show that the classification accuracy for the training and testing data are 80.31% and 80.05% respectively. This research demonstrates that ST is well suited for feature extraction since selected samples maintain the most important images of the original data and LS-SVM has great potential in classifying the EEG signals.