Student Performance and LMS Activity in Primary Schools: A Bayesian Additive Regression Trees Approach with Random Effects

Abstract Using data collected on almost all students aged 9 to 12 in Uruguay, we apply Bayesian Additive Regression Trees (BART) with random effects to study the association between student performance, Learning Management System (LMS) activity, and socioeconomic status. Performance data were combined with LMS activity pattern data.BART was chosen because of its strong predictive performance in high-dimensional problems. Additionally, it allows the inclusion of school-level random effects and, as a Bayesian method, provides an internal measure of uncertainty for its predictions.Results suggest that the model can be used for the early identification of at-risk students and to highlight schools that are either particularly successful or in need of intervention. LMS activity is characterized by several predictor variables, making it difficult to assess its overall effect on student performance. For example, some methods widely used in machine learning applications—such as variable importance measures and profile plots (e.g., Partial Dependence Plots)—are not suitable, as they are designed to evaluate only one or a few predictor variables. We address this limitation using a synthetic student profile approach to assess the effect of LMS activity on academic performance. An interesting finding is that high levels of LMS usage have greater positive effects on performance, particularly among students from low socioeconomic backgrounds.