Classification of confusion states of students viewing online classes using machine learning models

Massive open online classes (MOOC) had gradually become a popular way of learning, through which students could choose corresponding courses according to their own time, eliminating the space and time constraints in traditional education. However, in this teaching method, due to the lack of effective feedback from students, the teacher did not know whether the students were still confused, resulting in lower teaching efficiency. Previous studies had tried using eye movement tracking and electroencephalogram (EEG) to determine whether students were confused, which led to low accuracy. In the study, machine learning models were used to increase the accuracy. EEG signals were collected through a wireless single-channel Mindset device, and four machine learning models, namely Logistic Regression, Random Forest, Gradient Boosting Decision Trees, and Neural Network were applied to classify the confused states in the learning process of students. It was found that all four machine learning models had the ability to classify confused states of students with a high accuracy rate. Additionally, Gradient Boosting Decision Trees outperformed three other models with an ROC-AUC of 0.85. Moreover, it was found that “Gamma1”, “Gender”, and “Alpha2” were three features with strong correlation with classification confusion. The present study shed light on the future study of confused states for students participating in online learning platforms.