Predicting Drop-Out from Higher Education: Evidence from Italy

Predicting university dropout is crucial. Identifying at-risk students can inform dropout prevention policies,
safeguarding the nation’s resources and mitigating the long-term deterioration of human capital. In contrast
to previous literature, this study prioritizes predicting student dropout rather than delving into causal
mechanisms. This study leverages administrative data encompassing the entire population of Italian students
enrolled in bachelor’s degree programs for the academic year 2013–2014. Our quantitative findings indicate
that machine learning algorithms exhibit significant predictive capabilities, specifically random forest and
gradient boosting machines, underscoring their potential as early warning indicators. Feature importance
analysis emphasizes the role of students’ first-year academic performance in dropout prediction. Furthermore,
our findings provide additional evidence regarding the influence of family income, high school grades, and
high school type. The adoption of these novel predictive tools can facilitate the targeted implementation of
policies aimed at mitigating this phenomenon.