Application of the XGBoost Algorithm to Predict Insecticide-Treated Net Use and Identify Its Determinants Among Pregnant Women in 27 Sub-Saharan African Countries: Analysis of DHS Data (2016–2024)

Background The burden of maternal mortality in Sub-Saharan Africa remains a critical public health challenge, with pregnant women facing disproportionate risks to their health and survival. Maternal mortality, often exacerbated by preventable conditions like malaria, has far-reaching effects on families, communities, and national development. Insecticide-treated bed nets are a cornerstone of malaria prevention, yet utilization remains suboptimal due to multifaceted barriers. While traditional statistical methods have explored these factors, machine learning offers transformative potential to predict insecticide-treated bed nets use and uncover complex determinants through data-driven insights. This study addresses the gap in applying advanced machine learning techniques to optimize malaria prevention strategies in high-burden regions. Objective This study aimed to predict insecticide-treated bed nets utilization and identify its key determinants among pregnant women in 27 SSA countries using an XGBoost classifier and SHapley Additive exPlanations analysis. Method Nationally representative data from the Demographic and Health Surveys spanning 2016–2024 were analyzed, encompassing 32,279 pregnant women across 27 SSA countries. Data preprocessing addressed class imbalance via ADASYN, and Bayesian optimization tuned the XGBoost model. SHAP analysis elucidated feature importance, with Python 3.12 employed for model development and interpretation. Result The XGBoost classifier demonstrated robust performance, achieving an AUC of 93%, accuracy of 92%, precision of 93%, recall of 96%, and an F1 score of 95%. SHAP analysis identified rural residence, lower wealth quintiles, limited education, and inadequate healthcare access as dominant predictors of reduced insecticide-treated bed nets utilization. Mobile phone ownership and proximity to health facilities emerged as facilitators, while financial constraints and cultural barriers inhibited adoption. Conclusion and Recommendation: The XGBoost model effectively predicts insecticide-treated net utilization and highlights key determinants such as rural residence, socioeconomic status, education, and healthcare access. The model can identify high-risk groups, enabling more targeted ITN distribution. Policymakers should prioritize rural distribution networks, subsidized net programs, and mobile-based health education. Mobile health apps and AI-powered chatbots can provide personalized reminders and educational messages, addressing barriers to ITN use. This study emphasizes the potential of machine learning to guide targeted, equity-focused malaria interventions.