Predictive Modeling for Private University Enrollment Trends: A Study Based on Educational Fair Participant Behavior

Abstract

This study aimed to develop a predictive model for forecasting the likelihood of student applications to private universities. The specific objectives were 1) to analyze behavioral factors during open house events that are associated with application decisions; 2) to examine the relationship between participation behavior and application decisions; and 3) to develop a predictive model for application likelihood using behavioral data from open house participants. The dataset consisted of 1,321 registered participants, of whom 564 actually attended the event, 625 did not attend, and 132 students both registered and submitted applications during the event. The study employed Naive Bayes, Logistic Regression, Deep Learning, Decision Tree, and Random Forest models.

The key findings were as follows 1) the most important factors influencing event participation were the duration of attendance and the number of sub-activities visited; 2) actual participation behavior at the event was significantly associated with the decision to apply. The most influential factors in application decisions were duration of activity participation (Importance Score = 0.38) and number of sub-activities attended (Importance Score = 0.25), both reflecting levels of engagement and intention to apply; and 3) in developing a predictive model for application likelihood to private universities, the Deep Learning model achieved the highest performance (Accuracy = 0.94, AUC = 0.98), followed by the Random Forest model (Accuracy = 0.93, AUC = 0.96), which also demonstrated very high and comparable performance.

 The results highlight the capability of the Deep Learning model, based on artificial neural networks, to learn complex and non-linear behavioral patterns, such as the interaction between time spent at the event and the number of sub-activities attended, thereby outperforming traditional statistical models in predicting application decisions. Therefore, Deep Learning is concluded to be the most appropriate and effective model for use as a strategic decision-support tool in forecasting student admissions to private universities in this study.

The use of behavioral data processed for real-time visualization, in combination with machine learning techniques, can substantially enhance predictive accuracy and effectively support strategic planning for student recruitment in private higher education institutions.