Predicting STEM Achievement with Learning Management System Data: Prediction Modeling and a Test of an Early Warning System

Learning management systems log users’ behaviors, which can be used to predict achievement in a course. This paper examines the implications of data representations (e.g., dichotomous vs. count vs. principled, per learning theory) and applies forward selection algorithms to predict achievement in a biology course. Accuracy is compared across models. The paper closes with a description of an ongoing experiment that employs the prediction model, tests how multiple versions of an early alert message impact students’ access of learning resources, and compares the influence of messaging approaches related to personalization and feedback.