Predicting Students' Performance Using Machine Learning Algorithms
Charalampos Dervenis, Vasileios Kyriatzis, Spyros Stoufis, Panos Fitsilis
Abstract
Learning analytics (LA), defined as “the measurement, collection, analysis and reporting of data about learners and their contexts for the purposes of understanding and optimizing learning and the environments in which it occurs”, is a research topic that has gained significant importance and visibility among researchers during the past few decades. It is a research domain where the use of modern machine-learning (ML) algorithms and big data management provide timely and actionable information that can transform the overall learning experience for both students and educational institutions. In this paper we use ML algorithms in order to predict the performance of students, taking into account both past semester grades and socioeconomic factors. We run two models; a 2-class one predicting a “pass” or “fail” result and then we expanded this to a 5-class model, where we predict in which grading group the student will fall in the next semester. The results acquired indicate that it is possible to accurately predict the student's performance in both cases, with the 2-class model performing better than the 5-class one, which of course opts in providing more fine grain results.