Exploring Individual Feature Importance in Student Persistence Prediction

Abstract

Student persistence is of great importance for all stakeholders in higher education. There have been numerous studies using data mining and machine learning tools to predict student persistence. However, very little research has explored individual feature importance and their distinctive roles in predicting individual outcomes. In this study, we compare the predictive performance of two widely used machine learning models, logistic regression, and random forest, and use the SMOTE to improve the model performance. We analyze the feature importance in both aggregated form and their varied impact on individual predictions using a real-world student persistence dataset. In the discussion section, we propose practical approaches for monitoring and predicting student persistence.