Under-five mortality remains a critical public health concern in Sub-Saharan Africa (SSA), ranges from 59 to 230 per 1000 live births in different nations of SSA. Despite substantial efforts to reduce child mortality, progress in many SSA countries has been uneven and slow. Existing predictive models often fall short due to small sample sizes, and a lack of interpretability. Moreover, most studies have not leveraged large-scale, nationally representative datasets or applied robust machine learning (ML) frameworks. This study addresses these gaps by applying ML techniques to a large dataset to predict under-five mortality and identify key contributing factors. This study utilized the recent (2015–2024) Demographic and Health Survey (DHS) data from 23 SSA, incorporating a total weighted sample of 190,930 participants. About seven supervised ML algorithms were employed, applying data balancing, hyperparameter tuning, and model interpretability using SHAP values. Model performance was assessed using accuracy, precision, recall, F1-score, and AUC. This study results indicate random forest classifier demonstrated superior predictive performance (i.e., Accuracy = 0.9421, Precision = 0.9193, Recall = 0.9687, F1-score = 0.9433, and AUC = 0.9678) compared to other models. The SHAP waterfall plot provides a local explanation for a single high-risk case. Individual features such as poor healthcare access, younger maternal age at first birth (below 18 years), and poorest wealth status increased the predicted risk of under-five mortality, while protective features like belongs to maternal age 25–29 years reduce the risk of under-five mortality. This study demonstrated the effectiveness of ML models in predicting under-five mortality in SSA. Among the seven supervised classification algorithms applied, the random forest classifier exhibited the highest predictive performance. The use of SHAP values provided valuable insights into the most influential predictors, such as healthcare access, age at first birth and wealth status highlighting their significant role in child survival. By leveraging ML techniques, this research offers a data-driven approach to identifying at-risk children and informing targeted public health interventions.
Asnake et al. (Sun,) studied this question.