Discrimination of Decision Confidence Levels from EEG Signals

To explore the capability of utilizing electroencephalograms (EEGs) for the measurement of human decision confidence levels, this paper develops a new visual perceptual decision confidence experiment. In this experiment, a visual perceptual decision-making task is performed by 14 participants, and their EEG data are recorded. The problem of measuring decision confidence levels is considered to be a pattern classification task, and two pattern classifiers are trained with differential entropy (DE), power spectral density (PSD), differential asymmetry (DASM), rational asymmetry (RASM), and asymmetry (ASM) features extracted from multichannel EEG data. We compare the performance of these features and find that the DE feature performs better than the others for measuring levels of decision confidence. The experimental results indicate that EEG signals offer good capability for measuring human decision confidence levels. The best performance of our proposed method in measuring five levels of decision confidence reaches an accuracy of 49.14 % and F1-score of 45.07 %, and for the extreme levels of decision confidence, the recognition accuracy reaches 91.28 %, with an average F1-score of 88.92 %. Topographic maps are also used to depict the neural patterns of EEG signals, suggesting that the posterior parietal cortex and occipital cortex might be sensitive brain areas for indicating decision confidence.