Machine Learning in Admissions?: Use of Chi-Square Automatic Interaction Detection (CHAID) to Predict Matriculants to
John H Hollman1, David A Krause
1Dep. of Physical Medicine and Rehabilitation, Mayo Clinic School of Health Sciences, 200 First Street SW, Siebens 7 33-PPT, Rochester, MN 55905, USA. Tel 507-284-9547. Hollman.John@mayo.edu.
Purpose:
Machine learning algorithms provide methods by which patterns in admissions data may be discovered that predict admissions yields in education programs. We used a chi-square automatic interaction detection (CHAID) analysis to examine characteristics that predict applicants most likely to matriculate into a physical therapy program after being admitted.
Methods:
Data from applicants admitted to our physical therapy program from the 2015-2016 through 2021-2022 admissions cycles were evaluated (n=413). Variables included applicants' ages, grade point averages, graduate record examination (GRE) scores, admissions and behavioral interview scores, sex/gender, race/ethnicity, home state classification, undergraduate major classification, institutional classification, socioeconomic status, and first generation to college status. A CHAID algorithm identified which variables predicted matriculation after being admitted.
Results:
Overall, 47.2% of admitted applicants matriculated. The CHAID algorithm generated a 3-level model with 5 terminal nodes that classified matriculants with 64.9% accuracy. Applicants more likely to matriculate than to decline an admission offer included in-state applicants and White/Caucasian border-state/out-of-state applicants with GPAs below 3.65.
Discussion:
While findings are program-specific, the CHAID analysis provides a tool to analyze admissions data that admissions committees may use to analyze their admissions processes and outcomes.
