Determinants and Dynamics of COVID-19 Vaccine Hesitancy in University Students: A Machine Learning Analysis

Background: Booster vaccine hesitancy poses a challenge to sustained COVID-19 immunization even among individuals who accepted primary vaccination. This study examined associated factors and patterns of change in vaccine attitudes among university students in Ontario, Canada. Methods: A cross-sectional survey dataset was analyzed using validated psychometric scales to measure hesitancy toward primary and booster COVID-19 vaccination. Changes in hesitancy were operationalized as the continuous difference between booster and primary scores (ΔVH). Gradient Boosting and XGBoost regression models were fitted to estimate ΔVH from demographic characteristics (age, gender, socioeconomic status), vaccination history, and attitudinal constructs including complacency, confidence in vaccine safety, and perceived necessity of vaccination. Predictor contributions were assessed using SHapley Additive exPlanations, and Gaussian Mixture Modeling was employed to identify latent profiles among students with increased hesitancy. Results: A substantial proportion of students demonstrated higher hesitancy toward booster doses. Attitudinal factors, particularly complacency and safety perceptions, were the most influential predictors of increased hesitancy, whereas sociodemographic characteristics showed limited influence. Three distinct profiles of booster hesitancy were identified, reflecting heterogeneous patterns of vaccine attitudes and behaviors. Conclusions: These findings suggest that booster hesitancy in the study population is primarily associated with modifiable perceptions and can be effectively characterized using machine learning approaches that may inform targeted public health communication strategies.