Explainable machine learning framework for predicting childhood polio vaccination in Somalia: Comparative performance of multiple algorithms and SHAP-based interpretation

Background Polio remains a critical threat in fragile states where routine immunization systems are frequently disrupted. In Somalia, protracted conflict and nomadic mobility create significant hurdles for vaccine delivery. This study aimed to predict polio vaccination status (Polio3) among children under five years of age and identify socio-demographic predictors through an explainable machine learning framework. Methods Data from 1,527 children aged 0–59 months from the 2020 Somali Health and Demographic Survey was analyzed. Eight supervised algorithms were evaluated using an 80:20 train-test split executed prior to preprocessing to prevent data leakage. Synthetic Minority Over-sampling (SMOTE) was applied to the training set to address class imbalance. Model transparency was achieved through SHapley Additive exPlanations (SHAP). A survey-weighted multivariate logistic regression served as the baseline for performance comparison. Results The national prevalence of completed polio vaccination among the cohort was 55.73%. The Gaussian Naive Bayes model demonstrated the highest classification accuracy (66.6%), while the Random Forest model reported the highest discriminative power (AUC: 0.714). SHAP interpretation identified administrative region (Togdheer and Woqooyi Galbeed), child age (0–59 months), and administrative region, child age, and maternal education were the primary predictive determinants of uptake. Geographic factors significantly outweighed the predictive importance of household wealth and child gender. Conclusion The optimized ensemble models provide actionable risk-stratification tools for health authorities. Findings suggest that a transition toward residence-aware and region-specific strategies is necessary to address immunization gaps. The implementation of mobile clinics in high-risk geographic zones and the enhancement of maternal information networks is recommended to improve population immunity. This framework facilitates the systematic identification of under-vaccinated sub-populations to optimize resource allocation.