Scholarlink

<p>Previous literature on predicting university dropouts is mainly limited to institutional dropouts. Our paper extends this narrow framework by relying on administrative data on study trajectories, covering the entire Swiss higher education system. Using machine learning techniques, we predict university dropouts at the national level, show how these prediction results differ from those using a single university perspective, and provide prediction models for transfers to other higher education institutions. The results show that using only pre-enrollment data, we can correctly classify about 73% of all students, with an AUC of 79. By adding academic performance variables from consecutive semesters, we can correctly label about 88% of students after the 4th semester, with an AUC of 89. However, adding information on transfers to other Swiss universities instead of using information on single institutions only hardly improves the predictive performance of the models. It is, therefore, not surprising that in contrast to predicting university dropouts, the models perform poorly in predicting transfers to other higher education institutions.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Crossing individual university boundaries: a comprehensive approach to predicting dropouts in the higher education system

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年8月6日

查看全文
  • Johannes Berens,
  • Andrea Diem,
  • Luis Rumert,
  • Kerstin Schneider,
  • Stefan C. Wolter

摘要

Previous literature on predicting university dropouts is mainly limited to institutional dropouts. Our paper extends this narrow framework by relying on administrative data on study trajectories, covering the entire Swiss higher education system. Using machine learning techniques, we predict university dropouts at the national level, show how these prediction results differ from those using a single university perspective, and provide prediction models for transfers to other higher education institutions. The results show that using only pre-enrollment data, we can correctly classify about 73% of all students, with an AUC of 79. By adding academic performance variables from consecutive semesters, we can correctly label about 88% of students after the 4th semester, with an AUC of 89. However, adding information on transfers to other Swiss universities instead of using information on single institutions only hardly improves the predictive performance of the models. It is, therefore, not surprising that in contrast to predicting university dropouts, the models perform poorly in predicting transfers to other higher education institutions.