Predicting the Number of “Active” Students: A Method for Preventive University Management

Abstract

Dropout prediction is an important strategic instrument for universities. The Austrian academic system relies on “student activity” for university funding, defined as accumulating 16+ ECTS credits per study year. This study proposes a combined method of machine learning and ARIMA models, predicting the number of studies eligible for funding in the next study year. Data from the University of Graz between 2013/14 and 2020/21 was used for machine learning, and data from 2011/12 to 2020/21 was used as a base for the ARIMA models. Repeated predictions for the outcome years 2018/19 to 2021/22 yielded values of accuracy at .82, precision at .76, and recall at .73. The results showed deviations between <1% and 7% from the official values. Differences may be explained by the influence of the COVID-19 pandemic. This study offers a new approach to gaining information about future successful students, which is valuable for the implementation of preventive support structures.