Threads of Complexity : Lessons learnt from Predicting Student Failure through Discussion Forums’ Social-Temporal Dynamics

Abstract

Identifying students-at-risk of failing a course or dropping out of a program is a significant problem in the fields of Learning Analytics and Educational Data Mining. Improving their early detection is important for enabling higher education institutions to design and provide resources to better support students. In addition, learning is a dynamic process, where social interactions are crucial as learning is not completely an individual or static achievement. The identification of students-at-risk of failing or dropping out a course is generally an imbalanced problem, as grade distribution is affected by several elements, and failing students are not always fairly represented. This research focuses on exploring the extent to which network structure in online discussion forum interactions can inform student-at-risk predictions through node oversampling.