Using machine learning to understand physics graduate school admissions

Abstract

Among all of the first-year graduate students enrolled in doctoral-granting physics departments, the percentage of female and racial minority students has remained unchanged for the past 20 years. The current graduate program admissions process can create challenges for achieving diversity goals in physics. In this paper, we will investigate how the various aspects of a prospective student's application to a physics doctoral program affect the likelihood the applicant will be admitted. Admissions data was collected from a large, Midwestern public research university that has a decentralized admissions process and included applicants' undergraduate GPAs and institutions, research interests, and GRE scores. Because the collected data varied in scale, we used supervised machine learning algorithms to create models that predict who was admitted into the PhD program. We find that using only the applicant's undergraduate GPA and physics GRE score, we are able to predict with 75% accuracy who will be admitted to the program.