Data-Driven Approach for Student Engagement Modelling Based on Learning Behaviour

A great number of people around the world learn through MOOCs platforms offered by higher institutions of learning. Effective utilization of the platforms for self-directed learning will help to foster the United Nations’ agenda of inclusive and equitable education. However, low engagement and high attrition rates are common issues in the platforms. We believe that automatic modelling of student engagement levels based on their learning behaviour (access and interaction with various learning activities) will enable the construction of effective personalization methods that will improve student engagement in the platforms. In this study, statistical techniques were utilized to investigate the relationship between students' self-reported perception of their learning and their actual learning behaviour. Moreover, machine learning algorithms were applied to develop models for effective modelling of engagement levels in real-time. Our experiments with a de-identified dataset from Canvas Network open courses show promise in building predictive models that can automatically detect student engagement levels based on learning behaviour. The results revealed that significant differences in academic performance exist among different learner types and that machine learning models can be applied to automatically detect learners with low engagement. The models developed in this research can be applied to provide tools for instructors to observe student involvement in their learning activities so that they can devise pedagogical mechanisms/interventions to improve engagement through instructional design.