Early Predictors of Completion and Performance in a Gateway Mechanical Engineering Course: A Retrospective Cohort Study

Gateway engineering courses often expose accumulated gaps in students’ foundational knowledge, affecting progression and retention. This study investigates early predictors of completion and performance in Machine Elements, an integrative second-year mechanical engineering course at the University of Zagreb. A retrospective cohort dataset of 2351 students enrolled between 2013 and 2021 was analysed using admission variables, first-year course grades, and exam-attempt indicators. Cohort differences were examined with descriptive statistics and non-parametric tests, while course completion was modelled using logistic regression, LASSO-based predictor selection, and temporal validation. Students who completed the course differed most consistently from non-completers in Strength of Materials performance and prerequisite exam attempts. A reduced five-predictor model retained most of the full model’s predictive ability, achieving ROC AUC = 0.749 in the pooled sample and 0.726 in temporal validation. Intervention-oriented evaluation showed that Top-K risk selection provided high precision but low sensitivity under limited support capacity.