Towards analysing student failures: neural networks compared with regression analysis and multiple discriminant analysis

Using data from key first year courses, this article considers the development of subject-specific models to identify enrolled students at-risk of failure. The primary technique considered was neural networks, with itʹs results compared with logistic regression and multiple discriminant analysis. The three different modelling approaches were developed by three different analysts to achieve the benefits accruing from the independent M-Competition. We have found the quality of forecasts achieved to be significantly improved on earlier studies, presumably because of the subject specific nature of the models.