Predicting Bottle-Feeding Practices Among Mothers of Children Aged 0–24 Months in Somalia: A Machine Learning Analysis of the 2020 Demographic and Health Survey

Abstract Background In Somalia, where infectious diseases and malnutrition pose significant threats to child health, suboptimal infant feeding practices like bottle-feeding are a critical public health concern. This study aimed to determine the prevalence and identify key predictors of bottle-feeding among mothers of children aged 0–24 months using advanced machine learning approaches. Methodology: We analyzed data from the 2020 Somali Health and Demographic Survey (n = 5,416). Eight machine learning algorithms were employed to predict bottle-feeding status based on socioeconomic, demographic, and healthcare-related variables. Model performance was evaluated using accuracy, AUC-ROC, precision, recall, and specificity metrics. Results The prevalence of bottle-feeding was alarmingly high at 54.28%. Key predictors included household wealth (AOR = 0.66 for rich vs poor), place of delivery (AOR = 0.76 for facility vs home delivery), and child's sex (AOR = 1.13 for males). The Random Forest model demonstrated superior performance (Accuracy = 73.68%, AUC = 0.802), with geographic region, residence type, and parity emerging as the most important predictive features. Conclusion Bottle-feeding is remarkably prevalent in Somalia and strongly associated with poverty, limited healthcare access, and sociocultural factors. This practice contradicts WHO recommendations and exposes infants to substantial health risks in Somalia's challenging environment. Recommendations: Targeted interventions should focus on high-prevalence regions, integrate breastfeeding support with poverty reduction programs, improve access to health facilities, and address cultural beliefs through community education. Implementation of predictive models could enhance targeted public health efforts to promote optimal infant feeding practices.