Answers are grounded in this paper and cite their sources.
Under-five mortality remains a major public health concern in Somalia. Identifying its determinants and developing accurate predictive models is essential for guiding targeted interventions. This study aimed to identify factors associated with under-five mortality and to compare the predictive performance of logistic regression and machine learning models, using data from the 2020 Somalia Demographic and Health Survey (SDHS). A cross-sectional study was conducted using data from 13,438 children under 5 years. The outcome was under-five mortality (alive or dead). Descriptive statistics and Machine Learning were used to identify significant predictors, including Logistic Regression, Random Forest, LightGBM, and XGBoost models, which were developed and compared using accuracy, precision, recall, F1-score, and AUC-ROC. Feature importance was also assessed across machine learning models. Among the 13,438 observations, 466 (3.47%) were classified as deaths. Random Forest achieved the highest accuracy (83.24%), recall (52.17%), F1-score (0.127), and ROC-AUC (0.655), demonstrating the strongest overall discrimination among the evaluated models. Logistic Regression recorded the highest average precision (0.067), while XGBoost achieved the highest specificity (85.66%) and lowest Brier score (0.116). Preceding birth interval, place of residence, media exposure, region, and maternal age emerged as important predictive features across the models. However, low precision across all models indicated substantial difficulty in correctly identifying mortality cases. Random Forest demonstrated comparatively better predictive performance, but the modest discrimination and low precision of all models highlight the need for improved class-imbalance handling and external validation. These findings provide preliminary evidence for developing data-driven approaches to advance child-survival research and inform targeted maternal and child health interventions in Somalia. Not applicable.
No takes yet. Share an insight, caveat, or question.
Osman et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: