Student Retention Modelling: An Evaluation of Different Methods and their Impact on Prediction Results

Entering engineering students’ cognitive data from high school and their non-cognitive selfbeliefs can be influential factors affecting their academic success and retention decision. Effectively modelling the relationships between these early available factors and student’s future status of persistence in engineering can be particularly valuable to improve student retention in engineering. In this paper, twenty retention modelling systems were developed based on a combination of five retention models and four prominent modelling methodologies. These five retention models contain different collections of cognitive and/or non-cognitive factors, ranging from 9 to 71 input variables. The four modelling methodologies compared are: neural networks, logistic regression, discriminant analysis and structural equation modelling. Prediction performance results from these twenty modelling systems show that 1) neural network method produced the best prediction results among these four methods consistently, and 2) models combining both cognitive and non-cognitive data performed better than cognitive-only or noncognitiv-only models.