Application of machine learning techniques for predicting child mortality and identifying associated risk factors

Despite continuous persistent efforts to enhance child health through, among other things, universal access to care, child mortality remains a significant public health concern on a global scale. Child mortality is attributed to several factors including birth asphyxia/trauma, demographic and socioeconomic factors, preterm birth and intrapartum-related complications, pneumonia, preventable and treatable diseases, congenital anomalies, poor access to quality healthcare, poor hygiene and nutrition, and sanitation among others. In many sub-Saharan African nations, including Zimbabwe, the use of machine learning techniques to predict child mortality is still in its infancy. Therefore, this study applied machine learning algorithms decision trees, random forest, logistic regression and XGBoost to develop child mortality predictive models that utilize nationally representative demographic and health survey data. The logistic regression classifier achieved an accuracy of 74
^%
, random forest 72%, Decision tree 72%, and XGBoost a high accuracy of 81%. All under-five predictive models achieved a precision of 95 %. However, logistic regression achieved a recall of 76%, random forest 74
^%
, Decision tree 74
^%
, and XGBoost 84%. Logistic Regression achieved F1-score of 84%, random forest 83%, Decision tree 83% and 89% for XGBoost. The XGBoost outperformed other under-five predictive models. Integrating such models into health information systems can significantly assist policymakers and healthcare professionals to improve the health status of children, access to quality care and most importantly, improve preventive measures, immunization programmes, policies, and decision-making to improve child health. Understanding the risk factors can assist in designing intervention programmes aimed at improve child health while reducing child mortality.