Ranking of Determinants of Under-Five Mortality in Kenya Using Statistical and Machine Learning Approaches

Abstract

Under-five mortality (U5M) stands at 37 deaths in 1000 live births in Kenya suggesting that the country is unlikely to meet the World Health Organization target of fewer than 25 deaths per 1,000 live births by 2030 under current trends. While several determinants of U5M have been identified, evidence on their relative importance in Kenya based on nationally representative data and multiple analytical techniques, remains limited, constraining effective prioritization and use of scarce health resources. This study applied both traditional statistical methods and machine learning approaches to identify and rank the key determinants of U5M in Kenya. Data were obtained from the 2022 Kenya Demographic and Health Survey (KDHS), including 23,433 children aged 0–59 months. Feature ranking was conducted using chi-square tests, logistic regression, XGBoost, Boruta, and SHAP. To enhance robustness, results from the multiple methods were integrated using a heatmap-based consensus approach from which the average ranking of predictors across techniques was derived. Across all methods, maternal education consistently emerged as the most influential determinant of U5M, followed by maternal health status, household wealth index, ethnicity, and birth spacing. Literacy, ownership of household assets, and place of residence showed moderate importance, while the child’s sex was consistently ranked as the least influential factor. By integrating multiple statistical and machine learning techniques, this study provides robust evidence on the relative importance of U5M determinants in Kenya. Therefore, policymakers should prioritize investments in female education, maternal health, culturally responsive interventions, poverty reduction, and optimal birth spacing to accelerate progress toward achieving Sustainable Development Goal 3.2.