Predicting Student Attrition: A Case Study of the University of Namibia Bachelor of Accounting (Chartered Accountancy) Programme

Understanding the reasons behind student attrition is essential to developing retention strategies. In this study, we developed models to predict student attrition for the Chartered Accountancy programme at the University of Namibia (UNAM). Three hundred and eighty-nine records of accounting students were collected from the UNAM student database. We applied five classification machine learning algorithms: logistic regression (LR), support vector machines (SVM), k-nearest neighbors (k-NN), decision trees (DT) and Gaussian Naive Bayes (GNB) to identify patterns in student attrition. After, accuracy, specificity, sensitivity, F-measure and Area Under Curve were computed to measure the performance of each model. Finally, Shapley additive explanations (SHAP) was used to determine how variables influenced model prediction. The LR, k-NN, DT, SVM and GNB achieved an accuracy of 80.3%, 73.8%, 82%, 80.3% and 77%, respectively. Overall, the DT classifier performed the best at predicting student attrition. We found that academic performance in year 1 core modules: financial accounting A and financial accounting B were the most important features, on average contributing the most to the predicted probability of dropping out. We have demonstrated that it is feasible to predict students who are likely to abandon their studies while in their first year. The major difference between our study and previous studies is that we do not focus on traditional studied areas (science, technology, engineering and mathematics). In this study, we developed predictive models to identify student attrition in the bachelor of accounting program in a Namibian university.