An Ensemble Method to Predict Student Performance in an Online Math Learning Environment

ABSTRACT The number of e-learning platforms and blended learning environments is continuously increasing and has sparked a lot of research around improvements of educational processes. Here, the ability to accurately predict student performance plays a vital role. Previous studies commonly focused on the construction of predictors tailored to a formal course. In this paper we relax this constraint, leveraging domain knowledge and combining a knowledge graph representation with activity scopes based on sets of didactically feasible learning objectives. Specialized scope classifiers are then combined to an ensemble to robustly predict student performance on learning objectives independently of the student's individual learning setting. The final ensemble's accuracy trumps any single classifier tested