Identifying Students At-Risk with an Ensemble of Machine Learning Algorithms

Abstract

Accurately predicting the academic performance of students is an important process for the long term sustainability of higher education institutions (HEIs). A reliable and timely identification of students at-risk will enable HEIs to take proactive measures to assist these students. Owing to large student numbers with varying backgrounds, identifying students at-risk is often primarily a manual process involving an analysis of students' prior academic results and little else. While prior academic results do play an important role in indentifying students at-risk, there could be several other factors that may be overlooked. The demographic and financial circumstances of the student for instance could play a key role in more accurately identifying students at-risk. It may be impractical to consider all of these factors when manually trying to identify students at-risk. We propose that classification models using an ensemble of machine learning algorithms (MLAs) have promising prospects to more effectively predict students at-risk by including several factors that may often not even be practical with manual methods.