Clustering technique-based least square support vector machine for EEG signal classification





Introduction

The electroencephalogram (EEG) signal is a measure of the
summed activity of approximately 100 millions of neurons
lying in the vicinity of the recording electrodes [1]. It may provide
insight into the functional structure and dynamics of the
brain [2,3]. Basically the EEG is complex and aperiodic time
series containing information of the electrical activity generated
by cerebral cortex nerve cells. The EEG signal is generally
used as a diagnostic indicator for investigating brain activities
under various physiological conditions [4]. The recording
of EEG is one of the most important clinical tools for studying
the functional states of the brain and for diagnosing and
monitoring neurological diseases [5,6], such as epilepsy.


Previous research

Up to now, several techniques have been proposed for the
classification of EEG signals in the literature, and diverse classification
accuracies have been reported in the last decade
[9–22,27]. A brief description of the previous research follows.
Guo et al. [22] proposed the relative wavelet energy and an
artificial neural network for the classification of EEG signals.
They implemented their method on the epileptic EEG database
for healthy volunteers with eyes open and epileptic patients
during seizure activity. Their method attained a 95% classification
accuracy. In Jahankhani et al. [20], the researchers
used wavelet for feature extraction and neural networks for
EEG signal classification on the same set of the epileptic EEG
data and they achieved 98% classification accuracy. Subasi [18]
introduced a mixture of an expert model with a double-loop
Expectation-Maximization (EM) algorithm for the detection of
epileptic seizures.



EEG data

In this study, three databases are used to assess the method.
They are obtained from three different sources, one for the
epileptic EEG data [23] and one for the mental imagery tasks
EEG data [24] and one for the motor imagery EEG data [25]. The
three databases are briefly described below.
The epileptic EEG data, developed by the Department of
Epileptology, University of Bonn, California, and described in
[29], is publicly available [23]. The whole database consists of
five EEG data sets (denoted as A–E), each containing 100 singlechannel
EEG signals of 23.6 s from five separate classes. Each
signal was chosen after visual inspection for artefacts, such as
causes of muscle activities or eye movements.


Proposed methodology
In this paper, we propose a new algorithm of the CT-LS-SVM
for classifying EEG signals. The block diagram of the proposed
CT method based on LS-SVM for EEG signals classification is
shown in Fig. 4. The first block is the input of EEG brain signals
and the second block is the feature extraction