Machine Learning for Predicting Childhood Immunization Status and Identifying key Predictors Among Children Aged 12–23 Months in Sub-Saharan Africa: A Multi-Country DHS Analysis

Abstract Background Childhood immunization is one of the most cost-effective public health interventions for preventing vaccine-preventable morbidity and mortality; yet substantial coverage gaps persist across Sub-Saharan Africa. These gaps are driven by persistent socioeconomic inequities, geographical disparities, and barriers to the provision of consistent service delivery that impact immunization uptake across multiple levels of influence. Machine learning possesses the capability to identify high-risk children and the key factors leading to under-immunization by examining complex relationships within large population datasets. These studies aimed to development and evaluate standard Machine learning models for predicting of childhood immunization status and identifying key important predictors among children aged 12–23 months in Sub-Saharan Africa using multi-country DHS data. Methods A pooled cross-sectional analysis was performed utilizing DHS Kids Recode datasets from 22 Sub-Saharan African (SSA) nations, collected between 2019 and 2025, encompassing a total of 49,817 children aged 12 to 23 months. Seven machine learning algorithms were developed and assessed, including Logistic Regression, Decision Tree, Random Forest, K-Nearest Neighbors, Artificial Neural Network, XGBoost, and LightGBM, employing metrics such as ROC-AUC, PR-AUC, accuracy, precision, recall, F1-score, and calibration. To interpret model predictions and pinpoint significant factors influencing childhood immunization, SHapley Additive exPlanations (SHAP) were utilized. Results Among the machine learning models evaluated, the Random Forest algorithm exhibited the most superior predictive capability regarding childhood immunization status, attaining an accuracy of 94.24%, alongside a ROC-AUC of 0.9875 and a PR-AUC of 0.9890. The prevalence of complete immunization among children aged 12 to 23 months was found to be 40.11%. The interpretation of the model using SHAP highlighted the key factors that contribute to the predictions of childhood immunization status. The most significant predictors included adequate antenatal care (ANC) utilization (four or more visits), institutional delivery, postnatal care (PNC) receipt, maternal education (secondary and higher education), partner education (secondary and higher education), older maternal age, skilled birth attendance, and urban residency. These factors collectively emphasize the critical role of maternal healthcare engagement, educational empowerment, and equitable access to essential health services throughout the continuum of maternal and child care in influencing childhood immunization status. Conclusion Machine learning models have exhibited robust capabilities in predicting childhood immunization status and in identifying significant predictors of immunization outcomes. The integration of explainable artificial intelligence boosts the interpretative aspect of these models, thereby facilitating the development of targeted, data-driven strategies that aim to strengthen vaccination programs and improve childhood immunization rates in sub-Saharan Africa. Such advancements are crucial for optimizing public health initiatives and ensuring better health for children in the region.