Using Machine Learning Techniques and Matric Grades to Predict the Success of First Year University Students

Student enrolment and biographical data are rich sources of information that could help universities and staff tackle a diverse range of problems, such as identifying at risk students, student intake limitations, and course content adjustments. South Africa faces a unique economic and political history which creates new sets of challenges in the determination of which students are at risk of failing their degrees. This paper investigates which attributes of a student best predict whether they will graduate so as to identify vulnerable students and offer them crucial assistance. Different machine learning algorithms are applied to the data and the results are compared. The data was synthetically generated using a Bayesian network with features such as the major a student chooses, their school quintile, high school grades as well as NBT scores. Bagging produced the best results, correctly classifying 75.97% of the data.