The application of Machine Learning for Early Detection of At -Risk Learners in Massive Open Online Courses

With the rapid improvement of digital technology, Massive Open Online Courses (MOOCs) have emerged as powerful open educational learning platforms. MOOCs have been experiencing increased use and popularity in highly ranked universities in recent years. The opportunity to access high-quality courseware content within such platforms, while eliminating the burden of educational, financial and geographical obstacles has led to a growth in participant numbers. Despite the increasing participation in online courses, the low completion rate has raised major concerns in the literature. Identifying those students who are at-risk of dropping out could be a promising solution in solving the low completion rate in the online setting. Flagging at-risk students could assist the course instructors to bolster the struggling students and provide more learning resources. Although many prior studies have considered the dropout issue in the form of a sequence classification problem, such works only address a limited set of retention factors. They typically consider the learners’ activities as a sequence of weekly intervals, neglecting important learning trajectories. In this PhD thesis, my goal is to investigate retention factors. More specifically, the project seeks to explore the association of motivational trajectories, performance trajectories, engagement levels and latent engagement with the withdrawal rate. To achieve this goal, the first objective is to derive learners’ motivations based on Incentive Motivation theory. The Learning Analytic is utilised to classify student motivation into three main categories; Intrinsically motivated, Extrinsically motivated and Amotivation. Machine learning has been employed to detect the lack of motivation at early stages of the courses. The findings reveal that machine learning provides solutions that are capable of automatically identifying the students’ motivational status according to behaviourism theory. As the second and third objectives, three temporal dropout prediction models are proposed in this research work. The models provide dynamic assessment of the influence of the following factors; motivational trajectories, performance trajectories and latent engagement on students and the subsequent risk of them leaving the course. The models could assist the instructor in delivering more intensive intervention support to at-risk students. Supervised machine learning algorithms have been utilised in each model to identify the students who are in danger of dropping out in a timely manner. The results demonstrate that motivational trajectories and engagement levels are significant factors, which might influence the students’ withdrawal in online settings. On the other hand, the findings indicate that performance trajectories and latent engagement might not prevent students from completing online courses