PERFORMANCE DEGLI STUDENTI UNIVERSITARI: METODI DATA-DRIVEN PER IL SUPPORTO DECISIONALE

This article analyses longitudinal data concerning university students through statistical and machine learning models, with the aim of evaluating academic performance, dropout risk and educational pathways within higher education systems. The study is based on a real dataset provided by the University of Pavia, including 231,740 observations related to 53,726 students over the period 2018–2022. The research compares parametric approaches, such as Linear Mixed Models (LMM) and Generalized Linear Mixed Models (GLMM), with non-parametric ensemble methods, including RANDOM FOREST, ADABOOST and XGBOOST. The findings show that machine learning models, particularly RANDOM FOREST and XGBOOST, provide higher predictive performance in identifying students at risk of dropout and in forecasting academic outcomes. At the same time, parametric models maintain greater interpretability and explanatory capacity, which remain essential for institutional decision-making processes. The article highlights the relevance of data-driven approaches for supporting university governance and educational policies, emphasizing the need to balance predictive accuracy, interpretability and policy applicability. The results suggest that the integration between advanced quantitative methods and institutional knowledge can contribute to the development of more effective strategies aimed at improving student retention and academic success.