Student Yield Maximization Using Genetic Algorithm on a Predictive Enrollment Neural Network Model

Institutions that allocate scholarship effectively among their prospective students benefit from improved enrollments, improved retention and potential increase in state and federal funding. Accordingly, the primary objective of this research is to develop a scholarship distribution model that enables academic enrollment offices to maximize student yield through efficient scholarship distribution. This paper presents the design of and tests a multi-layer feed-forward neural network (NN) in modeling the student yield factor. For this model inputs are assumed to be ACT score, GPA/class-rank, EFC, FAFSA, zip code and scholarship award amount and the single output is the student yield, where a one/zero system for accepting/declining the offer in attending the university is considered. The network is trained by applying the back error propagation algorithm, and is tested on holdout samples. A reliable testing result of 80% is achieved for the trained student yield neural network model. Having this model in hand, an optimization technique, Genetic Algorithm (GA), is applied to find an optimum scholarship distribution that maximizes total student yield.