Analytical Framework to Understand the Under-Five Survival in Ghana: A Hybrid Modelling Framework

Abstract

This study develops an integrated, comprehensive analytical framework to clarify the most significant determinants of under-five mortality in Ghana. The novelty lies in the fact that the four modelling paradigms are combined in a single pipeline as logistic regression, multilevel logistic regression, generalised additive models (GAM), and explainable machine learning (XGBoost + SHAP) and is the combination of the risk structure of the linear, nonlinear, hierarchical risk and interaction-driven risks. The data comprised 34,663 children and 17 covariates, which were analysed to examine the effects of socioeconomic, demographic, environmental, and service-related variables on survival. In the approach to methods, each model exhibited good, repeatable discrimination (AUC = 0.760-0.764). The GAM smooth models revealed that mortality risk decreased non-linearly with paternal age. However, XGBoost explainability showed that paternal age, residence, household gender arrangement, water source, and maternal-newborn services used were the predominant predictors. Calibration diagnostics indicated a reasonable estimate of probabilities which can be applied in everyday decision-making in operations. The triangulated evidence points to the importance of service access in early life, access to clean water, and household gender relations in determining the chances of survival. By introducing methodological novelty into child survival analytics and offering policy-oriented implications to be undertaken in line with SDG 3, SDG 6, and SDG 10, the study offers an innovative methodological approach to this issue. The findings recommend that health authorities intensify attention on maternal-newborn and clean-water interventions in high-risk households to accelerate equitable gains in child survival.