Determinants and Predictive Modeling of Long-Acting Reversible Contraceptive Use in Sub-Saharan Africa: Evidence from DHS Data Using Machine Learning and Association Rule Mining

Background : Long-acting reversible contraceptives (LARCs), including intrauterine devices and implants, are highly effective in preventing unintended pregnancies. Despite their benefits, utilization remains low across many Sub-Saharan African (SSA) countries. Objective : This study aimed to predict LARC utilization and identify key determinants among women of reproductive age in SSA using advanced machine learning techniques. Methods : A secondary analysis was conducted using the latest Demographic and Health Survey (DHS) datasets from eight SSA countries, yielding a weighted sample of 29,016 women aged 15–49 years. Data preprocessing included cleaning, feature engineering, variable selection, and class balancing with SMOTE. Twelve machine learning models were developed, and the best-performing model was optimized using Bayesian methods. Association rule mining (Apriori algorithm) was applied to uncover hidden patterns among predictors. Results : The pooled prevalence of LARC use was 29% (95% CI: 21%–38%), with high between-country heterogeneity (I² = 99.67%). Random Forest achieved the best performance after optimization, with an accuracy of 87.1%, AUC of 81.0%, and F1 score of 85.0%. Major predictors included country, education, parity, marital status, and age. Association rule mining showed that rural, uneducated, poor, and married women in Senegal and Burkina Faso had a higher likelihood of LARC use (Lift = 2.25). Conclusion : Machine learning identifies potential predictors of LARC utilization and identifies key determinants in SSA. Targeted interventions focusing on rural, low-income, and low-education groups may improve LARC uptake and reduce unmet family planning needs.