Classification of EEG evoked in 2D and 3D virtual reality: traditional machine learning vs. deep learning

Abstract Backgrounds
Virtual reality (VR) simulates real-life events and scenarios, widely used in education, entertainment, and medicine. VR can be presented in two or three dimensions (2D or 3D), and 3D VR produces a more realistic and immersive experience. Previous research has revealed that the electroencephalogram (EEG) induced by 3D VR has a different profile from that of 2D VR, manifesting in many aspects, such as the power of brain rhythm, brain activation, and brain functional connectivity. However, studies on how to classify EEG in 2D and 3D VR were limited.
Methods
64-channel EEG was recorded, while visual stimuli were given in 2D and 3D VR. The classification of these recorded EEG signals was done using two machine learning methods: the traditional method and the deep learning method. In the traditional machine learning classification, EEG features of power spectral density (PSD) and common spatial patterns (CSP) were extracted, and three classification algorithms, support vector machines (SVM), K-nearest neighbors (KNN), and random forests (RF), were used. A specialized convolutional neural network, EEGNet, was used in the deep learning classification. These classification approaches were compared with respect to their classification performance.
Results
In aspects of four performance evaluations for classification, accuracy, precision, recall, and F1-score, respectively, classification using the deep learning method is better than the traditional machine learning approaches. Classification accuracy with deep learning with EEGNet could reach up to 97.86%.
Conclusions
The classification performance of 2D and 3D VR-induced EEG can be achieved with EEGNet-based deep learning, outperforming conventional machine learning approaches. Given the role of EEGNet, which is designed for EEG-based brain-computer interfaces (BCI), better performance classification of EEG in 2D and 3D VR environments might be predicted to be helpful for the application of 3D VR in BCI.