Term of Award

Fall 2023

Degree Name

Master of Science, Electrical Engineering

Document Type and Release Option

Thesis (restricted to Georgia Southern)

Copyright Statement / License for Reuse

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.


Department of Electrical and Computer Engineering

Committee Chair

Rocio Alba-Flore

Committee Member 1

Fernando Rios

Committee Member 2

Mohammad Ahad


When visual stimuli are presented at appropriate frequencies, the brain creates steady-state visual evoked potentials (SSVEP) that matches the brain activity. This study uses the non-invasive electroencephalography (EEG) device Emotiv EpocX to measure brain activity. The data for this research was obtained from IEEE data port, and the dataset comprises 20 SSVEP event recorders. 7Hz, 9Hz, 11Hz, and 13Hz were employed in the experiment, coupled with a baseline recording. Support Vector Machine was used to determine which of the frontal lobe channels to use. Six channels were initially selected for use in this project but were later reduced to four by using a Support Vector Machine (SVM). Therefore, only four of the Epoc X's 14 channels were used in this study; two channels from the frontal lobe and the two occipital channels, which are most suitable to detect emotions SSVEP EEG signals, were employed for this procedure. Power spectral density and discrete wavelet transform feature extraction techniques were used to extract features from each channel, and these features were used to train three separate algorithms to classify SSVEP EEG data. This thesis focuses on enhancing SSVEP EEG signal categorization algorithms and evaluating which Long-Short Term Memory (LSTM) neural network, Gated Recurrent Unit (GRU), and a Recurrent Neural Network (RNN) performs better. Also, to test the whole process, the method was tested using a second dataset obtained from IEEE data-port with events recorded at 5Hz, 6Hz, 7Hz, and 8Hz. It was observed that LSTM had a classification accuracy with 75 percent accuracy, 70 percent for GRU and 74 percent for RNN, when performing classification for the first dataset. Similarly, when performing classification on the second dataset, LSTM had a classification accuracy of 74 percent, GRU of 72 percent, and RNN of 75 percent.

OCLC Number


Research Data and Supplementary Material