ABSTRACT
The COVID-19 has resulted in schools to pause face-to-face classes in traditional classrooms and shifted to online classes in virtual classrooms across the globe. Unfortunately, students' ability to pay attention in class is uncertain. Hence, this study developed models that detect student attention in a virtual class through facial expression. In this study, for every 15-second video segment, statistical values such as the mean, median, and variance of Facial Action Units of each volunteer participant were extracted. Same video segments were labeled independently by three domain experts with attentive or inattentive to form the dataset. Such dataset was then split into 80-20 training-testing split ratios. Results showed that the model developed by the Decision Tree classifier using the Information Gain split criterion gave the best performance with an accuracy of 90.00% and a kappa of 0.796. The presented rate of the accuracy implies a high percentage of correctly predicted observations, while the high kappa value implies a very strong agreement between our human annotators and the machine in labeling students' level of attention. © 2022 IEEE.