EEG feature extraction strategies for personality classification

Determine whether using resting-state EEG features beyond power spectra—specifically oscillatory phase and temporal correlations across channels or source space—can increase the accuracy of machine learning classifiers in predicting Big Five personality trait scores and their lower-order aspects from resting-state EEG recordings.

Background

The study trained machine learning models on power spectral features from resting-state EEG recorded with 32 electrodes to predict Big Five personality traits and their ten lower-order aspects in 309 participants. Across extensive nested cross-validation and multiple preprocessing and modeling choices, classification and regression performance did not exceed chance or null baselines.

In the Discussion, the authors note that relying solely on power spectra may miss potentially informative EEG features. They suggest that oscillatory phase or temporal correlations in channel or source space could add information, raising the unresolved question of whether alternative feature extraction strategies would enable successful personality prediction from resting-state EEG.