Abstract
The purpose of this article was to identify factors that predict the dropout rate in 4-year universities using machine learning algorithms. The Random Forest and XGBoost algorithms were employed, and data from the Korean Council for University Education and the University Information Disclosure System were utilized for model development. The collected data spanned from 2011 to 2021 and included a total of 17 independent variables related to educational finance, educational conditions, and educational and research performance. The dropout rate was analyzed as the dependent variable. The final Random Forest and XGBoost models demonstrated the ability to explain 65% and 66% of the variability in the dropout rate, respectively. Using the two machine learning models, the top 10 explanatory variables were identified based on their importance, excluding the student enrollment rate. The models identified the following variables as important predictors of the dropout rate: expenditure per student for academic materials, research budget per full-time faculty member within the university, research budget per full-time faculty member outside the university, freshman enrollment rate, employee-to-student ratio, and employment rate. By examining the partial dependence plots of the five common important explanatory variables, it was analyzed how the predicted dropout rate changes in response to variations in these variables. Although there were differences in the curves between Random Forest and XGBoost, the overall direction and trends of the curves for the important explanatory variables were found to be similar. The improvement of these five important variables was found to have a negative correlation with the dropout rate, indicating a decrease in the predicted dropout rate. Based on these research findings, key implications and suggestions for further studies were provided.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have