Predicting Learners’ Performance in a Programming Massive Open Online Course

Abstract

In recent years, predictive models in Massive Open Online Courses (MOOs) have mostly focused on predicting student success in cohort MOOC environments which are designed with structured timing and planned content release. However, in self-taught courses, which are characterized by their flexibility in timing and release of content, predictions can be more critical because students’ success depends on their behavior during learning. Where, student behavior is defined by the combination of complex variables that describe their interactions with course resources. Therefore, existing models must be adapted in such a way as to consider heterogeneity in student behavior. To address this need, this paper studies how student interactions with self-taught MOOC resources can be included in predictive models. Twelve types of interactions with video-readings, assessments and supplements are analyzed to measure their effect on predicting success in a population of 38,838 students enrolled in a course. Additionally, this work contributes to a methodology that aims to improve predicative models of student performance in a course by identifying student profiles and their probability of success. Results of this work show that the interactions of students with a course have a high predictive power, among them the most relevant are completing video-readings, completing evaluations, and reviewing previously completed supplements.