Development of a Model that Detects Student's Disengagement during an Online Lecture Presentation through Eye Tracking and/or Head Movement
Andrea Ranaika Basinillo, Byron Matthew Oracion, Raphael Magno, Larry A. Vea · 2019
Technologies such as online lecture presentations are now provided to students to cater their learning through digital means. Unfortunately, student's engagement during online classes is uncertain. To partially address this problem, we developed models that can detect student disengagement using some well-known decision tree classifiers. Results showed that Random Forest using Information Gain criterion provided the highest accuracy rate of 91.73% and kappa statistic of 0.74. It was also observed that majority of the internal nodes of the tree models are features extracted from eye tracking. This includes: the time duration of the eye seen, the distance of the iris from the center of the eye, and the distance travelled by the iris. It implies that eye features are more likely to be determinants in detecting student disengagement rather than head movement features. Finally, we significantly noticed that: in every 450 video frames, if the eye is seen for less than 76 frames and an average angle of head pitch is less than or equal to 46.5 degrees with respect to the negative y-axis, the student is most likely disengaged. We suggest that the model should be embedded in a computational system that automatically provides feedback to teachers conducting online classes which could aid the teacher in maintaining the class' engagement and help pave way to future studies regarding any individuals' engagement.