Our paper aims to classify electroencephalography (EEG) data as part of the Brain-Computer Interaction (BCI) competition. We evaluate and compare the effectiveness of different deep learning architectures, including Vanilla Convolutional Neural Networks (CNNs), Deep Convolutional Networks (DeepConvNets), Shallow Convolutional Networks (ShallowConvNets), EEGNet, and hybrid architectures like CNN with Recurrent Neural Networks and Long Short-Term Memory (LSTM) (CNN-RNN-LSTM). Specifically, the architectures are tested on a single subject and with all subjects. Additionally, time duration variations in EEG data are also taken into consideration and compared. Our findings indicate that EEGNet achieved the highest test accuracy of 78%, demonstrating its potential for accurate EEG signal classification across many subjects.
More details are covered in "ECEC147 Report"
Dataset is made publicly available through: http://www.bbci.de/competition/iv/
Depending on which architecture you would like to use, the respective notebooks are attached in the GitHub Repository