Hybrid Convolutional Neural Networks for Multi-Emotion Classification Using GAMEEMO

Emotion classification is an important step in understanding human affective states, and it is also applied in fields like psychology, health care, and human–computer interaction. Electroencephalography (EEG) is a promising modality for objective emotion assessment because of the ability to capture neural activity associated with emotional processing. This research focuses on developing a data-driven approach for emotion classification using GAMEEMO EEG signals. In this study, emotion analysis was performed by proposing deep learning models. This work carried out in three phases. First, EEG data is preprocessed and analyzed using the MNE library where all brain frequencies were filtered. Secondly, building five different hybrid deep learning models using CNN is one of the first works that use the hybrid CNN-GRU model to classify the GAMEEMO database. In the last stage, multi-class classifications were made using five developed models which are CNN, CNN-SVM, CNN-RF, CNN-XGBoost, and CNN-GRU, and an accuracy of 93%, 91.26%, 91.36%, 91.21%, 95%, respectively, was obtained.