Advancing Knowledge on Machine Learning Algorithms for Predicting Childhood Vaccination Defaulters in Ghana: A Comparative Performance Analysis

High rates of childhood vaccination defaulting remain a significant barrier to achieving full vaccination coverage in sub-Saharan Africa, contributing to preventable morbidity and mortality. This study evaluated the utility of machine learning algorithms for predicting childhood vaccination defaulters in Ghana, addressing the limitations of traditional statistical methods when handling complex, high-dimensional health data. Using a merged dataset from two malaria vaccine pilot surveys, we engineered novel temporal features, including vaccination timing windows and birth seasonality. Six algorithms, namely logistic regression, support vector machine, random forest, gradient boosting machine, extreme gradient boosting, and artificial neural networks, were compared. Models were trained and validated on both original and synthetically balanced and augmented data. The results showed higher performance across the ensemble tree classifiers. The random forest and extreme gradient boosting models reported the highest F1 scores (0.92) and AUCs (0.95) on augmented unseen data. The key predictors identified include timely receipt of birth and week six vaccines, the child’s age, household wealth index, and maternal education. The findings demonstrate that robust machine learning frameworks, combined with temporal and contextual feature engineering, can improve defaulter risk prediction accuracy. Integrating such models into routine immunization programs could enable data-driven targeting of high-risk groups, supporting policymakers in strategies to close vaccination coverage gaps.