Research on Confusion Emotion Recognition based on Passive Brain-Computer Interface in Spatial Tasks

Confusion is a key manifestation of cognitive disequilibrium, and moderate confusion facilitates deep learning. Accordingly, recognizing confusion has become a core issue in learning emotion perception. Focusing on confusion recognition in spatial tasks, this paper proposed a confusion emotion recognition framework. On this basis, a passive brain-computer interface system based on electroencephalography (EEG) and its evaluation model were constructed, and the model accuracy was validated at both intra-subject and cross-subject levels. Traditional algorithms, including Naive Bayes, Support Vector Machine, and Random Forest, together with end-to-end approaches such as CNN and EEGNet, were employed to identify confusion states. Results showed that end-to-end approaches outperformed traditional ones in both binary and four-class classification tasks, while Random Forest achieved competitive performance close to end-to-end methods. Moreover, this paper further analyzed regional brain differences between confused and non-confused states, and found significant differences in the power spectral density of the frontal lobe, temporal lobe, parietal lobe and occipital lobe. This study established an experimental framework and performance baselines for EEG-based confusion recognition, and offered preliminary evidence for understanding brain regional differences associated with confusion in spatial tasks. It was expected to advance the intelligent recognition of learning confusion and provide references for subsequent educational affective computing applications.