Scholarlink

<p>Student dropout remains a critical challenge for higher education systems in socially and economically diverse regions. While artificial intelligence is increasingly used to predict dropout risk, limited attention has been paid to whether such systems reproduce or obscure underlying social inequities. This study proposes a counterfactual explainable AI framework to examine how ethnic and regional inequalities are encoded in predictive models of student dropout. Using longitudinal administrative data from a regional university in Vietnam’s Northern Midlands and Mountainous Area, the analysis evaluates three dimensions of algorithmic inequity: group-level risk disparities, under-identification of at-risk students, and counterfactual effort required to reduce predicted dropout risk below the model-defined intervention threshold. Results show that students from rural and ethnic minority backgrounds face higher predicted dropout risk, are less likely to be flagged by early-warning systems, and require systematically greater counterfactual effort to achieve comparable risk reductions. By interpreting counterfactual effort as a proxy for institutional burden, the study illustrates how explainable AI can function as a tool for social diagnosis rather than mere prediction. The findings highlight the importance of integrating explainability and fairness analysis to inform equity-oriented educational policy. Although situated in a Vietnamese context, this study does not claim direct empirical generalization to other higher education systems. The proposed framework is conceptually transferable to settings facing structural inequality, while the magnitude and patterns of disparity documented here should be interpreted as context-dependent.</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

Using explainable AI to diagnose institutional inequality in student dropout across ethnic and regional groups in Vietnam

  • 期刊论文

  • 发表日期：

    2026年4月19日

查看全文
  • Trung-Nghia Phung

摘要

Student dropout remains a critical challenge for higher education systems in socially and economically diverse regions. While artificial intelligence is increasingly used to predict dropout risk, limited attention has been paid to whether such systems reproduce or obscure underlying social inequities. This study proposes a counterfactual explainable AI framework to examine how ethnic and regional inequalities are encoded in predictive models of student dropout. Using longitudinal administrative data from a regional university in Vietnam’s Northern Midlands and Mountainous Area, the analysis evaluates three dimensions of algorithmic inequity: group-level risk disparities, under-identification of at-risk students, and counterfactual effort required to reduce predicted dropout risk below the model-defined intervention threshold. Results show that students from rural and ethnic minority backgrounds face higher predicted dropout risk, are less likely to be flagged by early-warning systems, and require systematically greater counterfactual effort to achieve comparable risk reductions. By interpreting counterfactual effort as a proxy for institutional burden, the study illustrates how explainable AI can function as a tool for social diagnosis rather than mere prediction. The findings highlight the importance of integrating explainability and fairness analysis to inform equity-oriented educational policy. Although situated in a Vietnamese context, this study does not claim direct empirical generalization to other higher education systems. The proposed framework is conceptually transferable to settings facing structural inequality, while the magnitude and patterns of disparity documented here should be interpreted as context-dependent.