Scholarlink

<p>Student dropout remains a persistent challenge for higher education institutions, particularly as student populations become increasingly diverse. Despite extensive research on retention, early identification of at-risk students and the implementation of timely, effective interventions remain difficult. This study applies machine learning methods to predict dropout risk using administrative data from 1,385 bachelor’s degree students enrolled in business and economics programs between 2014 and 2024. Several algorithms were tested, including logistic regression, gradient boosting, neural networks, and Random Forest. The Random Forest model outperformed the others, achieving an AUC of 0.961, an accuracy of 88.6%, and an F1 score of 88.5%. While Random Forest has been used in prior dropout studies, this research is novel in combining it with survival analysis and relying exclusively on basic enrolment data. The analysis identified place of residence and prior secondary school program as the strongest predictors of dropout. Age and citizenship also contributed, particularly in survival analysis, where younger and international students showed higher dropout risks. In contrast, study program and study mode had limited predictive power. The findings demonstrate that machine learning can function as an effective early warning system but should be complemented with qualitative evaluation to ensure ethical and supportive interventions.</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

Predicting student dropout in higher education using Random Forest and survival analysis

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月28日

查看全文
  • Valerij Dermol,
  • Vesna Skrbinjek

摘要

Student dropout remains a persistent challenge for higher education institutions, particularly as student populations become increasingly diverse. Despite extensive research on retention, early identification of at-risk students and the implementation of timely, effective interventions remain difficult. This study applies machine learning methods to predict dropout risk using administrative data from 1,385 bachelor’s degree students enrolled in business and economics programs between 2014 and 2024. Several algorithms were tested, including logistic regression, gradient boosting, neural networks, and Random Forest. The Random Forest model outperformed the others, achieving an AUC of 0.961, an accuracy of 88.6%, and an F1 score of 88.5%. While Random Forest has been used in prior dropout studies, this research is novel in combining it with survival analysis and relying exclusively on basic enrolment data. The analysis identified place of residence and prior secondary school program as the strongest predictors of dropout. Age and citizenship also contributed, particularly in survival analysis, where younger and international students showed higher dropout risks. In contrast, study program and study mode had limited predictive power. The findings demonstrate that machine learning can function as an effective early warning system but should be complemented with qualitative evaluation to ensure ethical and supportive interventions.