Emotion Recognition from EEG with Normalized Mutual Information and Convolutional Neural Network

• Main Authors: Maria, Mahfuza Akter, Akhand, M. A. H., Shimamura, Tetsuya
• Format: Conference Proceeding
• Language: English
• Subjects: Complexity theory; Computational modeling; convolutional neural network; Convolutional neural networks; Deep learning; Electroencephalography; emotion; Emotion recognition; Feature extraction; normalized mutual information
• ISSN: 2771-7917
• DOI: 10.1109/ICECE57408.2022.10088920

Emotion is the fundamental trait of human beings, and brain signals are prospectus for emotion recognition (ER). Electroencephalography (EEG) is a preferable brain signal for ER as it is non-invasive, fast, portable and easy to use. Automatic ER from EEG is a challenging computational intelligence or machine learning task due to the inherited complexity of EEG signals. The aim of this study is to analyze different sub-bands (e.g., Alpha, Beta, and Gamma) of the EEG signals to find the most appropriate band to classify emotions using deep learning. At a glance, this study investigated main EEG signals (i.e., full frequency spectrum) and its sub-bands for connectivity feature map (CFM) construction using normalized mutual information (NMI); and convolutional neural network (CNN) was used for emotion classification from NMI CFMs. Experimental results identified that NMI CFMs from Gamma band showed relatively better emotion classification accuracy by CNN than CFMs with other bands. Finally, the proposed ER method with NMI and CNN is revealed as a potential EEG based ER method showing better than or competitive to existing related methods.