Augmented EEG Signal Classification for Energy Data Visualizations

Classification of electroencephalography (EEG) data is dependent on data size, quality, and generalizability to decode invaluable information from brain-computer interface (BCI) systems. EEG data collection typically occurs within controlled laboratory environments using long sections. Participation in these experiments draws less attention due to the challenging nature of the EEG experiments and the intended objective. Hence, an approach that offers the potential for both cost and time efficiency is the use of artificial data generation. Data augmentation can improve the classification performance by increasing training data. This study employs a Generative Adversarial Network (GAN) model to enhance the empirically collected EEG dataset, which was introduced for the purpose of analyzing the perceptions of energy users regarding data visualizations. The data size is initially expanded through the use of a GAN model. Subsequently, the augmented data is assessed in terms of its classification performance. The findings demonstrate the outcomes of three distinct sample groups, specifically defined as male, female, and mixed. Each group is trained in the production of synthetic EEGs, separately. The results of the classification analysis indicate that the use of augmented data leads to improved performance in the scenario of creating spectrogram images and subsequently classifying them through the implementation of a Convolutional Neural Network (CNN). The results indicate that the female sample group exhibited the highest levels of performance in terms of valence and arousal, achieving accuracies of 87.9% and 89.06% respectively. This study presents promising results in terms of enhancing the size of EEG data and improving classification performance.