Investigation of Efficiency and Accuracy of Deep Learning Models and Features with Electroencephalogram (EEG) Data for Binary Classification

Electroencephalogram (EEG) data is regularly used with Machine Learning to further develop the Medical Technology field. This research investigates the effectiveness of the popular Deep Learning models with a binary classification problem with the subjects' EEG data. The dataset used for this research comprised of EEG data available online that was recorded from human subjects who were labelled as either alcoholic or control. Each subject was presented visual stimuli and brainwave data was recorded through 64 electrodes located on each subject's scalp. Three types of features were included in this study: Raw signal data, Discrete Wavelet Transform (DWT) and Continuous Wavelet Transform (CWT). The study investigated how several Neural Network models performed when trained with the different features. The models were Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU) and Temporal Convolutional Network (TCN). The Convolutional Neural Network performed the best with the highest overall accuracy and the best AUC Score when trained with Raw data and Discrete Wavelet Transform. The Temporal Convolutional Network yielded the best AUC Score when trained with Continuous Wavelet Transform.