Educational Data-mining to Determine Student Success at Higher Education Institutions

Abstract

The expansion of enrolments in South African higher education institutions has not been accompanied by a proportional increase in the percentage of students who graduate. This is an ongoing problem faced by the Department of Higher Education and Training in South Africa (DHET). In their 2020 undergraduate cohort studies, DHET reported that the percentage of first time entering students graduating in minimum allocated time from 3 year degrees has remained low, ranging between 25.7% and 32.2%, for the academic years 2000 to 2017. This indicates students are struggling in higher education, as more than 60% of students being admitted by the system are consistently not completing their chosen field of study in the allotted time. In this study, we introduce an approach that involves prediction of student performance at each year of study until qualifying, for students at a South African higher education institution. The present study applies various classification techniques to a synthetic data-set, generated by a Bayesian network, with the aim to show that these classifiers can be used to predict student performance in advance with the aim to promote student success and avoid the negative consequences of students struggling to complete their studies or dropping-out altogether.