No abstract

It is of great importance to detect users’ confusion in a variety of situations such as orientation, reasoning, learning, and memorization. Confusion affects our ability to make decisions and can lower our cognitive ability. This study examines whether a confusion recognition model based on EEG features, recorded on cognitive ability tests, can be used to detect three levels (low, medium, high) of confusion. This study also addresses the extraction of additional features relevant to classification. We compare the performance of the K-nearest neighbors (KNN), support vector memory (SVM), and long short-term memory (LSTM) models. Results suggest that confusion can be efficiently recognized with EEG signals (78.6% accuracy in detecting a confused/unconfused state and 68.0% accuracy in predicting the level of confusion). Implications for educational situations are discussed.