Modeling Graduate Student Success Utilizing Admissions Data: A Pilot Study

Abstract

Assessing student success is critical in university graduate school admissions. This is inherently difficult due to the heterogeneous applicant pool, widely varying grading standards across international universities, and other factors. This research is a pilot study to develop mathematical models that leverage admissions data to predict student success in graduate school for a limited set of graduate programs. The goal is a repeatable workflow to support (not replace) graduate admissions committees. Admissions data is matched with academic performance in key courses selected by Directors of Graduate Programs (DGPs). Initial supervised learning models include random forest decision trees and regression to analyze the data and produce useful models, including program-specific scoring models for international school-degree pairs. Initial data exploration led to the construction of multiple models to answer several key questions from DGPs regarding program admissions policies. The findings will highlight factors that contribute to student success and frame potential decisions regarding the admissions process, curriculum changes, and support to improve the likelihood of student success. This proof of concept is a foundation for future research in the study of graduate student success. This research provides a workflow to exploit additional data, DGP feedback, and additional models to support robust decision-making regarding admissions and support for graduate students.