Predicting university dropout: connecting big data and structural models
研究对比了基于大数据的预测模型与基于学生个人变量（投入度、满意度）的结构方程模型在预测大学生辍学上的结果，发现两者一致性有限，强调变量选择和预测分析权重的重要性。
There is almost universal and long-standing concern regarding the high dropout rates among university students. Determining the causes in order to reduce the risk of dropout has been a recurrent research topic. Interactive-causal models, based on structural equations (SEM), have recently been joined by other procedures based on data mining or academic analytics. The aim of this work was to analyse the convergence between a predictive model on academic dropout based on big data and the results of a structural equation model (PLS-SEM) defined on the basis of the student’s personal variables (engagement and satisfaction) that previous research has shown to be highly relevant. The results confirm the relationships between the main variables and dropout probability, mediated by academic performance. However, the limited agreement between the prediction methods highlights the importance of carefully selecting variables and weighting predictive analyses. This is crucial to avoid overestimating dropout likelihood or adopting overly deterministic approaches that overlook the relational and interactive aspects of the issue.