DCGAN based EEG data augmentation in cognitive state recognition

Deep Learning (DL) has recently been applied to classify cognitive states of human operators for various activities based on their brain activity observed via electroencephalogram (EEG). However, we often face an underfitting problem when training DL-based classification models using EEG signals because of an insufficient amount of training data. To solve such a data scarcity problem, we propose a novel data augmentation framework for EEG using a Deep Convolutional Generative Adversarial Network (DCGAN) to improve the performance of a Convolutional Neural Network (CNN) classifier for motor imagery tasks. We evaluated the proposed DCGAN-based EEG data augmentation method with different amounts of data augmentation using the BCI competition dataset. The results showed that the classifier applying the proposed EEG data augmentation had higher classification accuracy in terms of all possible amounts of data augmentation except for only one case (subject “ay” with 25% data augmentation) when compared to the classifier without EEG data augmentation. Also, the classifier with 100% data augmentation had the highest classification performance as compared to others (0%, 25%, 50%, 75%). These results indicated that the proposed DCGAN-based EEG data augmentation framework could be an applicable method for improving the performance of a CNN classifier for cognitive states.