Predicting completion of massive open online course (MOOC) assignments from video viewing behavior

Abstract

ABSTRACT Predicting student performance in Massive Open Online Courses (MOOCs) is important to aid in retention efforts. Researchers have demonstrated that video watching features can be used to accurately predict student test performance on video quizzes employing neural networks to predict video test grades from viewing behavior including video searching (ff, rw, pause), replays, stop, and start. Deep learning neural networks are susceptible to overfitting with low data and higher dimensions; hence, we compare various commonly used classification algorithms including logistic regression and demonstrate similar or higher rates of prediction. However, using a path analysis approach we find that the features collectively explain only a small to moderate amount of variance in assignment completion, which suggests that other factors than video-viewing behavior influence assignment completion such as student goal motivation and student self-regulation. Overall, our findings highlight the important contribution of active searching and repeated viewing to successful assignment completion in a MOOC course. Predictive models based on user interactions with the MOOC platform can help target course retention strategies to increase MOOC completion where retention is abysmally low and help to target video viewing strategies to optimize teaching and learning platform functionality using adaptive agents.