Predicting factors that affect student performance in MOOC and on-campus computer science education

Abstract

We are analyzing the performance of students in both Massive Open Online Courses (MOOCs) and in residential courses. To do so, we are analyzing data gathered on two edX programming courses, as well as the residential counterparts for those courses that are administered in person at MIT. A large part of the research performed involved building out a data analysis educational software platform on which students taking the residential version of the course at MIT perform all of their work. Using this software, we gather data on students analogous to the data gathered on the MOOC students. Using this data, we will search for behaviors that lead to better performance in the courses. In addition, we use machine learning algorithms in order to be able to create predictive models that can determine how a student will perform in a course given their behaviors and background information. Finally, a major contribution of this paper is applying those machine learning models to the course software in order to provide the students with counseling as to which behaviors leads to increased performance and learning for students of their backgrounds.