OSU professor develops models for predicting collegiate freshmen retention

Thursday, September 2, 2010

Recent research conducted in Oklahoma State University’s Department of Management Science and Information Systems shows it may be possible to predict and prevent at-risk freshmen students from dropping out of college.

In two studies published in the Journal of Student Retention and Decision Support Systems, OSU-Tulsa researcher Dursun Delen created decision models that accurately predict which students are more likely to drop out at the end of their first year of college.

“Student attrition has become one of the most challenging problems for decision-makers in academic institutions,” said Delen, an associate professor of management information systems in the Spears School of Business. “In spite of all of the programs and services to help retain students, according to the U.S. Department of Education, Center for Educational Statistics, only about half of those who enter higher education actually earn a bachelor’s degree. The idea behind this research is to identify the very nature of collegiate attrition, especially at the freshmen level.”

Delen used a data mining methodology called Cross Industry Standard Process for Data Mining (CRISP-DM) to examine five years of anonymous OSU student records and develop analytical models capable of predicting at-risk freshmen with an accuracy of approximately 80 percent.

Delen used the models to analyze 39 variables ranging from demographic information, to students’ social interaction, students’ prior expectation from educations endeavors, and parents’ educational and financial background.

The studies revealed the most important predictor of freshmen student attrition was the number hours of coursework the students earned compared to the number of hours in which the student was enrolled. The more hours of coursework the students earned compared to the number enrolled, the more likely they would be to continue their education.

Delen also identified three other predictors as significant. In descending order of importance, these included whether or not the students had loans for the spring, how high their fall grade point average was, and whether or not they had a grant, tuition waiver or scholarship for the spring semester.

The practical implications of this study are twofold, Delen said. First, the studies show that academic institutions can pair Delen’s prediction models with data from existing databases to accurately identify at-risk students and optimize resources to retain them. Second, the prediction models can provide insight about which variables are the main determinant of student attrition at specific academic institutions.

In addition to predicting at-risk freshmen students, Delen also is working to identify which variables are most important in retaining sophomore, junior and senior students through graduation.

“I want to show the educational institutions they can do something about freshmen attrition beyond just talking about it,” Delen said. “As an educator, you want to retain them, you want to educate them, and you want to graduate every single one of them. So, at the end of every student’s first semester, whatever it is we know about those students can go into the models to predict which ones are more likely to drop out so we can do something about retaining them.”