Applying machine learning-based models to prevent University student dropouts

Abstract

In this paper, we explored models with good performance indexes for predicting student characteristics and dropout status to prevent students from dropping out. As a result of applying 6 classification models to 30,118 academic data of University A from 2018 to 2022, the accuracy rate of XGboost algorithm was 96.9% and the recall rate was 94.4%. XGboost was selected as the final model and the importance of the dropout influencing factors was high in the following order: total number of grade changes, number of semesters completed, number of leaves of absence, grade point average, grade level, and number of academic warnings. Finally, we proposed long-term and short-term management strategies for students with a high probability of dropping out of school through a consistent dropout prediction process.