Abstract:

This study investigates the determinants and temporal patterns of student dropout in Mathematics 1, an introductory, high-stakes course taken by all first-year students across engineering programmes at the Faculty of Engineering, University of Rijeka. Given the course’s foundational role and its consistently high attrition rates, understanding not only whether students disengage but also when dropout occurs is essential for designing timely and effective instructional interventions. To address this challenge, the study employs survival analysis as a rigorous analytical framework for modelling dropout dynamics throughout the semester.

Three groups of predictors are included:
(1) prior educational background, encompassing secondary school type, prior grades, and mathematics preparation;
(2) motivational and psychological characteristics collected through validated instruments administered at the start of the semester; and
(3) learning analytics extracted from the institutional LMS, capturing patterns of student behaviour such as frequency and timing of activity, engagement with course materials, and assignment-submission behaviour. Kaplan–Meier estimators are used to visualise survival functions and compare dropout trajectories across student subgroups, while Cox proportional hazards regression identifies key predictors associated with elevated or reduced dropout risk.

The results highlight distinct temporal windows in which dropout is most likely to occur and reveal a combination of academic, motivational, and behavioural variables that significantly influence the hazard of disengagement. These insights provide an empirical foundation for developing early-warning systems and targeted support mechanisms within STEM contexts. The study demonstrates how integrating learning analytics with survival modelling can strengthen institutional capacity to detect at-risk students and intervene at strategically meaningful moments in the learning process.

Keywords:

Survival analysis, learning analytics, dropout prediction, Kaplan–Meier, Cox regression, engineering education, Mathematics 1.