Predicting student teachers’ dropout with machine learning: potentials of using student and study progress data from the campus management system

Abstract

The unmet need for teachers draws attention to student teachers’ dropout. In this context, interest is both focused on clarifying the reasons and on reducing dropouts in teacher education programmes. With regard to the analysis of dropouts, new possibilities for the use of student and study progress data (learning analytics) are emerging in the context of digitalisation. The paper presents a study in which dropout predictions were calculated based on campus management data of 4601 student teachers. Two machine learning methods, logistic regression and random forest, were used and their application and results are presented. With both methods, all students could be correctly assigned to either the group of successful graduates or dropouts with about 80% accuracy. The most important predictor was the examination performance in the first three semesters (grade and percentage of passed examinations). The article discusses possibilities and challenges of dropout prediction in teacher education programmes as well as implications of data use.