Deriving financial aid optimization models from admissions data

Abstract

This paper presents a novel approach to deriving probabilistic models that predict enrollment given applicant background and the amount of financial aid offered. Our Bayesian network models can be used to optimize various enrollment objectives. We present a novel efficient optimization algorithm that uses the models to maximize expected tuition revenue under capacity constraints including student-faculty ratio and accommodation. We demonstrate and evaluate our approach using four years of graduate admissions data from the Asian Institute of Technology, consisting of 7,788 applicants from 84 different countries. This data set is particularly challenging since reliable family income data is not available for students from most of these countries. Evaluating the Bayesian network model with 10-fold cross validation yields an ROC Az value of 0.8451, with a predictive accuracy of 82.70% at a threshold of 0.5. Comparing the results of the tuition revenue optimization model to the institute's current financial aid allocation practice shows that if single-term tuition revenue is the sole optimization criterion, the institute can achieve its current enrollment numbers while realizing significant savings in its financial aid budget. The prediction and optimization software is currently being incorporated into the institute's online admissions processing system.