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

Abstract

Abstract. 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.

A pairwise output coding method for multi-class EEG classification of a self-induced BCI

“…As the accuracy, ease of use, efficiency and speed are important parameters to consider [26], the feature extraction approach proposed in [27] is used in this study. This method relies on the band powers of EEG signal which is a common and powerful technique to distinguish different frequencies [28][29][30]. There, a stable pattern in the PSD was observed with different amplitudes for all subjects and for all tasks.…”

“…Then the first feature is selected as the highest PSD peak value in the alpha band (8-13 Hz), which is referred to as f1 in the Figure 3. The second and third features are the arbitrary first and second highest PSD peak values in beta band (13)(14)(15)(16)(17)(18)(19)(20)(21)(22)(23)(24)(25)(26)(27)(28)(29)(30) , which are referred to as f2 and f3 in Figure 3. Proposed feature extraction scheme.…”

EEG signal classification based on simple random sampling technique with least square support vector machine

“…An overall classification of 95.96% was reported in their paper. At the same time, Siuly et al (2009) applied sampling techniques for feature extraction and LS-SVMs for EEG signal classification on the same data sets.…”

Classification of real and imaginary hand movements for a BCI design

“…The method proposed in this study relies on the frequency distributions of the signal's power spectral density. Most of the researchers have studied with epileptic data or data taken during hypnosis [16,17,[20][21][22][23], however this study use spectral analysis for classification of motor imagery tasks. The frequency characteristics are important for detection of mental and motor changes [4], so augmented power spectral density estimation is chosen as the feature extraction method for this study similar to study in [24].…”

“…The power spectral density (PSD) is used as the first feature extractor which is a powerful and common way of discriminating different frequencies [5,21,23]. PSD of the signal was calculated by using Welch Periodogram.…”