At-Risk Students Prediction Using Machine Learning

Abstract

This article intends to discover how machine learning can be used to predict at-risk students during the school year. Different algorithms were tested within a common framework to compare their accuracy and their interpretability. Using some education expert knowledge, we examined each model relevance in relation to the most important features they used. Attendance, language proficiency and interim test completion were found to be very deterministic in the models prediction capabilities; not a surprise but a validation of the adequacy of the technology for this difficult task.