In this paper, an analysis is presented of the enrollment funnel for prospective graduate school students, predicting application submission and enrollment. Efficient university outreach is critical to optimizing a positive interaction cadence for prospective students, reducing costs, and strengthening academic program revenue streams. Models employing rules, decision lists, and tree-based algorithms assessed the impact of a prospect's characterization and university outreach methods on application and enrollment probabilities. A novel two-stage modeling workflow, applied to each prediction problem, mirrored steps taken by a prospective candidate to become a future enrollee. This analysis could help a graduate school decide what communication mechanisms would be best utilized. Compared models were Gradient Boosted Machines (GBM), XGBoost, Random Forest, and Certifiably Optimal Rule Lists (CORELS). Application submission prediction using XGBoost achieved an ROC of 99.7%, while predicting enrollment with GBM had an ROC of 92.3%. The analysis concluded that there was insufficient university outreach activity to convert admitted applicants to enrollees. The modeling results support the notion that features related to prospect demographics and prospect-driven actions were more important as predictors for application submission and enrollment than the available university outreach activity volumes included in this analysis.
Merritt, Stephen; Francomano, Anne; and Garcia, Martin
"Optimizing the Enrollment Funnel with Decision Trees and Rule Based List,"
SMU Data Science Review: Vol. 3:
1, Article 3.
Available at: https://scholar.smu.edu/datasciencereview/vol3/iss1/3
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License