Go to AIMS 2025 Workshop T2P homepagePredicting Childhood Routine Immunization Status in Ethiopia Using Ensemble Machine Learning Algorithms
The study aims to develop a predictive model for assessing childhood immunization status in Ethiopia using ensemble machine learning techniques.The research follows an experimental approach. Data from the Ethiopian Demographic and Health Survey (EDHS), collected at five-year intervals, was preprocessed for quality assurance. The study employed several ensemble machine learning algorithms, including Extreme Gradient Boosting (XGBoost), CatBoost, Random Forest (RF), and Gradient Boosting, with a One-Versus-Rest class decomposition method. A total of 35,512 instances with 18 features were used, with an 80/20 training/testing dataset split. The models were evaluated based on accuracy, with XGBoost achieving the highest performance at 88.30%, followed by RF (87.17%), CatBoost (86.92%), and Gradient Boosting (84.16%). SHAP (Shapley Additive Explanations) values were used to identify the most significant factors influencing immunization status, including child’s age, region, mother’s occupation, current parity, and mother’s age.Using XGBoost and SHAP values extracted decision rules based on feature importance. These rules reveal specific patterns related to immunization status, providing evidence-based insights for policymakers.XGBoost was selected as the best predictive model for childhood immunization status in Ethiopia. The study highlights key factors for targeted vaccination programs and scheduling, emphasizing the importance of early immunization and maternal characteristics in improving child vaccination coverage.