Abstract
Depending on the impact of emotions on a person's performance and emotional disorders that can be the main cause of many mental illnesses, as well as the desire of technology to design machines that are able to change their performance according to a person's emotional states, the study of electroencephalography (EEG) signals to analyze the different dimensions of human emotions has become increasingly significant. Based on machine learning models, this study was designed to identify the five emotions of relaxation, happiness, motivation, sadness and fear using EEG signal analysis. EEG data were collected from 23 male master's students at Tabriz University, aged 24-31, as they watched five videos designed to elicit different emotional responses. After preprocessing to remove noise and artifacts, we extracted statistical and frequency-domain features from the raw signal. The features were labeled and selected using statistical tests. In the final step, five different emotions were classified using decision tree, linear discriminant analysis (LDA), Naive Bayes, support vector machine (SVM), K-nearest neighbor (KNN), ensemble, logistic regression and neural network. It has been verified that ensemble and decision tree models had the highest accuracy with 95.38% and 91.77%.