Predicting likelihood of enrollment among applicants to the UVa undergraduate program

Universities face substantial uncertainty when trying to obtain their target number of incoming freshman students. The type and number of students that are accepted introduce variability for schools. This paper presents a method for reducing the uncertainty of how many undergraduate students will enroll at the University of Virginia. In addition, certain subpopulations of the accepted student pool were analyzed to help predict the number of enrolled students from those sub-groups. To shed light on how to reduce this uncertainty, we analyzed the history of applicants to UVa and identified characteristics that make a student more or less likely to enroll. Logistic regression, neural network, and classification and regression trees models are used to predict whether each accepted student will enroll. The results indicate that the logistic regression model best predicts total undergraduate enrollment. The logistic model delivered the lowest yield error, predicting an enrollment of 3,658 non-waitlist students when 3,662 actually enrolled.