AN APPLICATION OF MACHINE LEARNING TO COLLEGE ADMISSIONS: THE SUMMER MELT PROBLEM

Abstract

We present educational data mining research, specifically an application of machine learning to college admissions. Meeting targets for the number of students admitted and enrolled is crucial for many institutions of higher education since tuition-based income often serves as a major component of the operating revenue budget. Enrollment targets and diversity goals can be hampered by summer melt: the phenomenon in which students who, after being admitted and having committed to attend a college or university in the spring, do not actually enroll in the fall. Using 6 years of data from 2014 through 2019 of students admitted to a small liberal arts college in California, we investigate the application of supervised machine learning models to predict and identify those admitted students who will decline their admission offers, those that will accept their admission offers, and those students who are in danger of "melting away" over the summer. Institutions can use our summer melt model to estimate how many and identify which students will fail to enroll in order to implement activities and provide support to achieve their enrollment goals. The results of our research should encourage other institutions of higher education to apply machine learning algorithms to their admissions data to effectively estimate the size of the incoming student body and achieve other institutional goals.