A similarity Based Algorithm for Predicting Academic Success in First-year Undergraduates

Abstract

According to the European Commission, “Early school leaving is linked to unemployment, social exclusion, poverty, and poor health” (European Commission, 2024). Due to the importance of reducing school drop-out rates, several authors have analyzed this phenomenon. Before the boom of Artificial Intelligence, instead of using or implementing programming methods, researchers applied written formulas and rudimentary graphical visualizations to predict academic completion and the main factors behind it; besides, the development of Machine Learning (ML) algorithms has enhanced the precision and performance of an ample variety of investigations including the educational field. In this paper, we use a dataset of undergraduate-level students at the Universidad Nacional Autónoma de Mexico (UNAM) to predict timely academic completion. We use seven common ML algorithms and propose a novel algorithm based on students' similarities according to the most relevant features in common. This algorithm shows a higher precision than some traditional categorical ML algorithms. This innovative way to predict academic success can support educators, pedagogues, and policymakers make better decisions at UNAM.