Identification of Neural Correlates of Confusion during Online Learning Using Electrophysiological Recordings and Machine Learning Modeling

In recent years, Massive Open Online Courses (MOOCs) face the challenge of lacking immediate feedback. But it is difficult for teachers to judge how well the students understand the class in the online environment. To solve this problem, this study adopted a single-channel electroencephalogram (EEG) device to monitor students' learning status and identified confusing signals through machine learning algorithms. First, this study validated the effectiveness of EEG signals in distinguishing between confused and non-confused states using the NeuroSky MindSet, and further explored the importance of different brain oscillations in confusion with machine learning modeling. It was found that CatBoost showed the best performance of predicting confusion. In addition, the three most important features were Gamma2, Beta2 and Gamma1. Future research can further improve the detection accuracy by expanding the data scale, optimizing the algorithm, and combining more behavioral indicators (such as eye movements, facial expressions, etc.), and finally realize the development of personalized learning assistance systems to help teachers dynamically adjust teaching strategies and optimize the effect of knowledge transfer.