Neural network in an artificial intelligence model for realization of affective computing based on electroencephalogram analysis

This paper analyzes the possibility of assessing the emotional state of a person by analyzing features of their brain activity using non-invasive neuro-computer interfaces. The analysis of recent publications dealing with the use of electroencephalogram (EEG) signals for assessing the emotional state is carried out and topical problems in this area are revealed. The main approaches to brain stimulation for obtaining informative EEG signals, as well as describing methods for their analysis and recognition. The architecture of a deep convolutional neural network for EEG data analysis is proposed, as well as a neural network model for classifying 4 emotions (fear, happiness, sadness, calmness) according to the Russell valency-arozality scale based on two convolutional neural networks. An experiment was conducted with 50 participants who watched emotion-laden videos. EEG data from 50 subjects were collected and used to train and test the neural network model. The results showed a high emotion classification accuracy of 94%±3.4% while using a wireless neural interface.