Profiling and visualizing student performance in MOOCs to enable adaptation of course material

Abstract

MOOCs on platforms such as Coursera often attract orders of magnitude more students than a typical classroom course and usually lack significant in-classroom interaction; therefore, it becomes much more difficult for instructors to “get to know” the class, especially when the class comes from diverse academic backgrounds. This research focuses on a specific course on Coursera, implementing several approaches to profiling student performance on in-video quizzes, including least squares regression with baseline predictor bias and Hamming distance minimization. A reasonably accurate 10-dimensional model was built for each student. A visualization method was developed that lets instructors quickly gain intuition for how students are performing on quizzes. Student performance was also profiled over time to determine trends in performance over the course. Finally, a method was developed to analyze quizzes based on student performance in order to find any that are too easy, difficult, or confusing. The results of this data analysis can be used to adapt course material in various ways - across all students, for various clusters, or for individual students.