Classification of Emotions via EEG Signals by Deep Learning Approach

Emotion arises from the human brain’s response to various external factors. The complexity and variability of human emotions in real life make the study of emotion recognition crucial for practical applications. Traditional machine learning methods, however, face a significant challenge due to the time-intensive and expert-dependent feature extraction process. To address this issue, end-to-end deep learning techniques have emerged as a promising alternative, leveraging the inherent characteristics of raw signals and their time-frequency distributions. Our research explored the application of convolutional neural networks (CNNs), long short-term memory (LSTM) networks, and a combined approach in recognizing emotions through EEG data, utilizing the renowned DEAP dataset for experimentation. The performance of the CNN and CNN-LSTM models in recognizing emotions from EEG signals was impressive, achieving accuracy rates of 89.11% and 91.47% for raw data extraction, respectively. Moreover, reducing the parameters by approximately 50% in each iteration, specifically in terms of dropout probability, proved to be effective.