Exploiting EEG to identify confused students in online learning: A machine learning model comparison

Abstract. The COVID-19 pandemic has forced most schools, including universities, to adopt online courses for teaching. Due to the breaking of the constraints of time and space, online courses still occupy a large proportion in education until now, and large-scale open online education courses will become a major trend in the future. Of course, problems also arise in this process. As teachers cannot receive feedback from students anytime and anywhere in online courses, such as micro expressions, body movements, etc., they cannot know whether students are confused. Therefore, how to effectively detect students' learning status has grown in popularity. In the last few years, machine learning has developed rapidly, and a large number of artificial intelligence have emerged, making it possible to detect whether students are confused by combining electroencephalography with machine learning. In order to determine whether students were confused, this study used a wireless single channel Mindset device to collect Electroencephalography (EEG) signals. Six machine learning modelsrandom forests, eXtreme Gradient boosting, K-nearest neighbors, gradient boosting machines, logistic regression, and support vector machineswere then chosen. The findings indicate that other machine learning models have a high accuracy rate in classifying students' bewilderment, with the exception of the logistic regression model. With a 99.69% accuracy rate, the eXtreme gradient boosting model performs better than a number of other models.