Predictive Video Analytics in Online Courses: a Systematic Literature Review

Abstract

The purpose of this study was to investigate the use of predictive video analytics in online courses in the literature. A systematic literature review was performed based on a hybrid search strategy that included both database searching and backward snowballing. In total, 77 related publications published between 2011 and April 2023 were identified. The findings revealed an increase in the number of publications on predictive video analytics since 2016. In the majority of studies, edX and Coursera platforms were used to collect learners' video interaction data. In addition, computer science was shown to be the top course domain, whilst data collection from a single course was found to be the most common. The results related to input measures showed that pause, play, backward, and forward were the top in-video interactions, whilst video transcript and subtitle were the least used. Learner performance and dropout were the primary output measures, whereas learning variables such as engagement, satisfaction, and motivation were investigated in only a few studies. Furthermore, most of the studies utilized data related to forums, navigation, and exams in addition to video data. The top algorithms used were Support Vector Machine, Random Forest, Logistic Regression, and Recurrent Neural Networks, with Random Forest and Recurrent Neural Networks being two rising algorithms in recent years. The top three evaluation metrics used were Accuracy, Area Under the Curve, and F1 Score. The findings of this study may be used to aid effective learning design and guide future research.