Abstract
The prediction of students' academic progression and early identification of at-risk students have been considered as essential factors contributing to the improvement of student retention rates for higher education institutions in Australia. This study is concerned with the development of Bayesian neural networks (BNNs) for predicting students' academic progression in the first semester of their studies. By using data extracted from 353 international undergraduate students who enrolled in full-time studying load at Sydney Institute of Higher Education (SIHE), an artificial neural network was developed and trained by the Bayesian inference framework to explore the ability of BNNs in the application of predicting students' academic progression. With the classification results of 85% Sensitivity and 70% Specificity on the training set, and 81% Sensitivity and 74% Specificity on the testing test, it is indicated that the developed prediction model based on BNNs can successfully predict the student's progression. Additionally, it is demonstrated in this study that the better generalizability and the ability to quantify uncertainty of predictions tend to be the key advantages of BNNs over standard neural networks in the application of predicting students' academic progression.