Detection of Student Drowsiness Using Ensemble Regression Trees in Online Learning During a COVID-19 Pandemic

Abstract

Online lectures are mandatory to deal with the implementation of education during the COVID-19 pandemic. This significant change certainly creates a different experience for students. Regarding online learning, several public health experts and ophthalmologists say that residual radiation from electronic screens is causing an epidemic of eye fatigue. Research on smart classrooms actually appeared several years ago, but in reality it has not been implemented according to the planned concept. The current smart classroom research environment only uses outdated methods, which make the computer system incongruent (such as decision trees in video feeds) or only to the level of empirical studies or blueprints, which are not much help for other academic footing or reference materials. to students. This study aims to build an intelligent system that can evaluate students' attention during online classes, use teaching videos as learning feeds and input for predictions and also use advanced algorithms in several computational domains, namely face segmentation, landmarking, PERCLOS observations, Yawning and decision analysis using Ensemble Regression Trees to detect students' sleepiness, which is expected to patch up the shortcomings of the PERCLOS algorithm and the problems found in the single regression tree-based implementation. Based on the results of the tests that have been carried out, the system developed has been able to observe sleepy objects in learning videos with an accuracy of 80% so that later it can be a lesson for teachers why there are students who are sleepy during online classes either because of uninteresting material or other reasons.