Predicting student dropouts with machine learning: An empirical study in Finnish higher education
Matti Vaarma, Hongxiu Li
Abstract
This study uses three machine learning models to predict student dropouts based on students' transcript, demographic, and learning management system (LMS) data from a Finnish university. The contribution of this research lies in 1) comparing the relative importance of LMS (Moodle) data with transcript and demographic data in degree program dropout prediction, 2) examining the predictive importance of different data features monthly as a function of time from enrollment, hence extending the prior end-of-semester research to a midsemester analysis, and 3) measuring the prediction performance of the models monthly. The results identify “accumulated credits” (transcript) the “number of failed courses” (transcript), and “Moodle activity count” (LMS) as the most important features, suggesting LMS has significant predictive power and should be considered alongside transcript and demographic data when predicting degree program dropouts. Moreover, we visualize how these factors' importance and prediction performance vary over time, revealing general longitudinal trends and fluctuations within semesters. Finally, we elaborate upon this study's contributions before highlighting its limitations.