Extraversion prediction from EEG coherence during a face-to-face interaction task using machine learning techniques

Researchers have begun investigating personality assessments using brain-imaging techniques, such as electroencephalography (EEG). However, previous studies usually utilised EEG power, resting state, and video stimulus in the extraversion classification study, which could be the factors contributing to insufficient accuracy. Thus, this study proposes to classify extraversion using EEG coherence during a face-to-face interaction task. A total of 32 healthy male individuals were selected for this study based on their scores on the Big Five Inventory (BFI) and the Eysenck Personality Inventory (EPI). Sixteen of the individuals were identified as extraverts, whereas the remaining sixteen were identified as introverts. The study employed the Kruskal-Wallis H test to identify the high-ranking features. For the extraversion classification, optimizable KNN and SVM were utilised, along with leave-one-out cross-validation. The findings indicated that employing 1624 EEG coherence features yielded an accuracy of less than 80%. However, when applying feature selection, the accuracy increased up to 84.4%. Hence, we believe the study offers valuable insights for extraversion classification.