Interpretative Machine Learning Model to Predict Prospective Student Enrollment Probability and Support Targeted Marketing for Higher Education Admissions

Private universities face increasingly intense competition in attracting new students, while their marketing resources for student recruitment remain limited. Low conversion rates from prospective leads to final enrollment pose challenges for admissions teams in prioritizing leads, regions, and marketing channels. This study develops an interpretable machine learning model to predict the enrollment probability of each prospective student as a decision-support tool for targeted marketing using public datasets from Kaggle. The dataset covers student recruitment records, demographics, marketing activities, and final enrollment status. Features include lead sources, preferred study programs, educational backgrounds or high school majors, regions of origin, and relevant derived features. Logistic Regression, Random Forest, and XGBoost models are compared. Class imbalance is addressed through resampling and class weighting techniques. Model performance is evaluated using PR-AUC, recall, cumulative gain, and the Brier score for probability calibration, while SHAP analysis is employed to identify key factors influencing enrollment. This study aims to identify the best-performing predictive model, determine core enrollment drivers, and optimize targeted marketing strategies sustainably for higher education institutions.