Publication Details

Abstract

The pursuit of a master’s degree is a significant academic endeavor, one that is influenced by a complex interplay of factors extending beyond traditional academic performance. In this study, we estimate the determinants of timely master’s degree completion (i.e., within 3 years) using modern machine learning models such as random forest, decision tree, extreme gradient boosting, gradient boosting, and AdaBoost. After analyzing 15 years of master’s cohort data from the University of Texas at San Antonio, a large, public, Hispanic-serving university, our findings indicated that gradient boosting with hyperparameter tuning was a reasonably superior machine learning model for predicting master’s degree completion at our institution. The selected model accurately predicted more than 80% of the cases in the study and demonstrated superior predictive performance compared to the traditional logistic regression model. In support of nontraditional student retention theory, the model identified that students with higher GPAs, younger students, full-time students, and students who took out student loans were more likely to graduate within 3 years than students with lower GPAs, older students, part-time students, and students without loans, respectively. Furthermore, demographic-structural components, which are often overlooked in machine learning models, proved to be important: students in departments with a larger number of faculty and higher representation of female and non- White faculty members had a greater likelihood of completing their master’s degree successfully. 

Keywords: master’s students, 3-year completion, machine learning, gradient boosting

Authors:

• Fikrewold Bitew
• Lauren Apgar

Acknowledgements

We would like to thank the Institutional Research and Analysis team at the University of Texas at San Antonio for their feedback and suggestions. Thank you to Khoi To for his insightful comments on an earlier draft of this manuscript. 

Copyright © Association for Institutional Research 2025