Student:	Nurhan GÜRSEL ÖZMEN
Supervisor:	Assoc. Prof. Dr. Levent GÜMÜŞEL
Department:	Makine Mühendisliği
Institution:	Graduate School of Natural and Applied Sciences
University:	Karadeniz Technical University, Turkey

Title of the Thesis:	Analysis and Classification of EEG Signals Recorded During Different Mental Tasks for a Brain Computer Interface Design
Level:	Ph.D.
Acceptance Date:	17/9/2010
Number of Pages:	172
Registration Number:	Di799

Summary:

In this study, features of EEG signals which are recorded during different mental and

motor tasks are extracted and the performances of those features are tested with two

different classifiers. EEG signals recorded during mental tasks are analyzed so that it is

aimed to help people with neurophysiological disorders. The tasks are relax state, problem

solving, imagination of right hand, imagination of left hand and imagination of the letter A.

A Biosemi ActiveTwo System is used to record EEG signals and the data gathered

from the chosen nine electrode channels are transferred to the MATLAB for analysis and

then purified from noise and outliers. After that a two step feature extraction algorithm is

applied. The first step involves calculating power spectral densities by Welch method, and

the second step is developed from this power spectral density data, which is determining

alpha and beta band characteristics as features. The extracted feature vectors are

transferred to Linear Discriminant Analysis and Support Vector Machines classifiers and

they are used for two class and multiclass classification. Classification results are obtained

for each channel and by using those results the most active channels for each task are

determined.

Among the classifiers, Support Vector Machines performed better than the Linear

Discriminant Analysis for both classification schemes. In addition the use of four

electrodes found enough for analysis. The achieved classification results showed the

eligibility of five different tasks for Brain Computer Interface applications. Finally, the

classification performances are considered by classification accuracy, sensitivity and

specificity constraints.

Key Words: EEG signal analysis, Brain Computer Interface, Feature extraction,

Classification, LDA, SVM.