Quantification and Discriminant Evaluation of Unpleasant Emotions by Frequency Analysis Using EEG and ECG

The purpose of this study is to verify whether machine learning using electroencephalogram (EEG) and electrocardiogram (ECG) as inputs improves accuracy. The participants, 16 healthy adults, were given two stimulations: resting and unpleasant stimuli. Their EEG was measured 180 s immediately after the stimuli. The beta band of EEG and LF, HF, and LF/HF of ECG were calculated. The accuracy of the neural network was then compared using an EEG-only, ECG-only, and combined EEG and ECG dataset. The accuracy of the neural network using the combined EEG and ECG dataset was 79.51%, which was higher than that of the other datasets. The results suggest that emotional responses to resting and unpleasant stimuli may differ in brain and autonomic nervous system activity, and that combining both EEG and ECG indices may allow for more accurate discrimination.