11/16/2021 - Self-learning early warning system to counter high dropout rates in higher educationLIfBi

When students drop out of their studies, considerable costs are incurred - not only for the students themselves, but also for universities and society. Reducing dropout rates is difficult, however, because German universities largely lack appropriate forecasting options and intervention measures. In the first LIfBi Lecture of the current winter semester, Prof. Dr. Kerstin Schneider, Chair of Finance and Taxation at the University of Wuppertal, presented four projects that build on each other and have developed an early warning system to predict impending dropouts and are investigating how the information thus obtained can be used.

In order to support students at risk of dropping out of their studies in good time, the use of early warning systems can be useful. Under the leadership of Kerstin Schneider, the research project "FragSte" developed such an early detection system for both a state university and a private university of applied sciences. The system uses administrative study data and determines the dropout probability of students with machine learning methods.

Early indicators of dropouts

Using only data collected at enrollment - such as student age and type of higher education credential - it was possible to correctly predicted 77% of all dropouts at the state university. In addition, the activity level of students in the first few weeks of theirIm their first semester of study is a reliable indicator of whether they will complete their studies - in part because dropout rates are particularly high at the beginning of a degree program.

Supporting students at risk

However, early detection alone is not enough. Once students at risk of dropping out have been successfully identified, two questions arise. First, do these students even want to be supported, or are they only enrolled pro forma? This problem arises in a special form in Germany, where there are no tuition fees and there are significant financial benefits for students. Second, what is the best way to intervene? As part of a field study, a research team investigated how interventions can affect students' dropout behavior. For example, students at risk were notified early and, in some cases, invited to counseling sessions. This had a positive effect on certain groups. However, it must be taken into account that students are a heterogeneous group and react differently to interventions, for example, depending on their field of study.

The potential of the NEPS

Finally, Kerstin Schneider discussed her project team's plan to design and evaluate an AI-based student support system and to focus on the role of student-subject matching in dropouts. She also emphasized the potential of combining the administrative data used in her projects with data from the National Educational Panel Study (NEPS), which is housed at LIfBi. Thus, NEPS data could provide information on the reasons for dropouts and not only open up a deeper understanding of the complex of issues, but also provide the basis for policy recommendations.

Link to the profile page of Prof. Dr. Kerstin Schneider at the Bergische Universität Wuppertal

Self-learning early warning system to counter high dropout rates in higher education

More News