Learning generalised features from EEG using deep learning for a cognitive brain-computer interface

• Main Author: Ang, Nigel Wei Jun
• Other Authors: Guan Cuntai
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Computer science and engineering
• Online Access: https://hdl.handle.net/10356/148292
• Institution: Nanyang Technological University

Attention can be measured by different types of cognitive tasks. Such tasks include the Stroop Test, Eriksen Flanker Test and the Psychomotor Vigilance Test (PVT). Despite all three cognitive tasks' different contents, they all require the use of visual attention. To learn the generalised representations from the Electroencephalography (EEG) data of different cognitive attention tasks, Intra-task and Inter-task attention classification experiments were conducted on three types of attention task data using SVM, EEGNet, and TSception. Via Intra-task experiments with Cross Validation, TSception has achieved significantly higher classification accuracies than other methods, with 82.48%, 88.22%, and 87.31% for Stroop, Flanker and PVT tests respectively. For the Inter-task experiments, Deep Learning methods showed superior performance over the SVM with most of the accuracy drops not being statistically significant (p>0.05). Our experiments indicate that there is common hidden knowledge that exists across data from the different cognitive attention tasks, and that Deep Learning methods can learn generalised representations of the data better than the SVM.