As financial pressures increase across higher education, researchers are turning to machine learning to better predict which universities are at risk of closure.
In a recent study, higher education researchers worked with the Federal Reserve System to develop a predictive model that combined hundreds of institutional characteristics to estimate the likelihood that a college would close. This model performs better than the financial oversight system currently used by the federal government and provides a more nuanced understanding of higher education institutions' financial distress.
Reasons why universities close
Although university closures remain relatively rare, economic hardship is more widespread. Smaller institutions, especially those that rely heavily on tuition revenue, have little margin for error when enrollment declines. Zach Mabel, director of research at Georgetown University's Center on the Education Workforce (CEW), which studies higher education finance, said universities could be in financial trouble for years if they don't close, and that most stressed institutions do not eventually close. This makes it especially difficult to predict which schools will close.
“[I] There may be a variety of factors why universities close, but there is no silver bullet, he said.
Philip Levine, an economics professor at Wellesley College, said the closures should be understood as the most extreme result of broader financial pressures across higher education. For example, higher education institutions rely heavily on human infrastructure, and their budgets are continually under pressure from rising labor costs that outpace inflation. Inflation-adjusted tuition rates have remained flat or declined over the past decade, putting educational institutions under continued operational stress. If a university is unable to reduce costs without compromising its functionality, the only remaining option is closure.
“You can lay people off,” he says, “and you can cut costs, but then someone has to stand in front of the room and teach the kids.” “And children need food and support and other services. And at some point it becomes unsustainable.”
A closure could have a devastating impact on students, employees, and the surrounding community that relied on this facility to boost the local economy.
Current method
Currently, the federal government's primary tool for assessing the financial health of colleges and universities is the U.S. Department of Education's Financial Responsibility Composite Score (FRCS). It is a metric constructed from three accounting ratios that measure liquidity, capital, and profitability. Institutions that fail to meet the composite score or fall into the warning range could face increased funding oversight, and universities would be required to disburse financial aid to students before being reimbursed.
These measures act as compliance and risk management mechanisms rather than predictive tools. They are based on historical financial data and use a limited number of pre-specified indicators. It is also based on information from non-standardized institutions, as not all schools use the same fiscal year.
Robert Kerchen, dean of educational leadership and policy studies at the University of Tennessee, Knoxville, and a contributor to the model, said the lack of timely data limits the usefulness of these measures. Federal data is based on audited financial reports and submissions to the Integrated Higher Education Data System (IPEDS). This means that financial stress becomes visible only after the federal system has been building for many years. For example, IPEDS data for fiscal year 2024 has just become available, he said.
machine learning
Machine learning models can address this problem by incorporating many variables at the same time and allowing the data to determine how those variables interact. Rather than assuming which financial metrics are most important, these models test a combination of enrollment trends, revenue sources, institution size, sector, and other characteristics to estimate closure risk.
In the researchers' model, declining enrollment and reliance on tuition revenue emerged as particularly strong predictors of closure. Other factors such as organization size, sector (public, private, not-for-profit or for-profit), and even missing data also contributed to the risk assessment. The model was tested against historical closure events and demonstrated higher predictive accuracy than FRCS-based monitoring alone, with an average accuracy of 83 percent compared to 77 percent for the federal method.
The advantage of machine learning, Mabel said, is not only that it can process more data, but also that researchers don't have to specify in advance how different elements relate to each other.
“The model itself will be looking at and testing all the different ways that multiple types of information can work together to provide a more comprehensive and holistic picture of the likelihood of closing a deal,” he said.
