Personality traits classification from EEG signals using EEGNet

Personality represents the individual differences in cognition and behavior. The five personality traits, as identified by the Big Five system, are traditionally assessed by using self-report questionnaires that are subject to bias problems. For this reason, the need for an automatic personality assessment method has emerged. Assessing personality from EEG signals recorded as a response to specific stimuli has shown promising results. In this work, we adopted the EEGNet, a compact CNN model developed for EEG decoding, to implement an automatic personality trait binary classifier. For this purpose, we used the EEG traces of the AMIGOS dataset, which were acquired on 38 subjects during the visualization of emotional videos. Different types of data preprocessing and different model hyperparameters were tested. The best performing model achieves classification accuracy of 0.93 for Agreeableness, 0.92 for Extroversion, 0.90 for Conscientiousness, 0.89 for Emotional Stability and 0.89 for Openness.