Predicting contraceptive need and identifying associated patterns among postpartum women in Sub-Saharan Africa: A machine learning and association rule mining study

Abstract

Background Unmet need for postpartum family planning (PPFP) remains a major public health challenge in Sub-Saharan Africa (SSA). Traditional regression models often fail to capture complex, non-linear interactions among predictors. This study developed an explainable machine learning framework combined with association rule mining to predict postpartum unmet need and extract policy-relevant multi-factor risk profiles. Methods We analyzed pooled Demographic and Health Survey (DHS) data from 66,454 postpartum women across 27 SSA countries (2016–2024). Logistic Regression, Random Forest, and XGBoost models were optimized using stratified five-fold cross-validation. SMOTE was applied strictly to training data to resolve class imbalance. Model performance was evaluated on an independent 20% test set using accuracy, precision, recall, F1-score, and AUC-ROC. SHapley Additive exPlanations (SHAP) prioritized key features for Apriori-based association rule mining (thresholds: Support , Confidence , Lift ). Results Overall unmet need for PPFP was 24.0%. Random Forest achieved the highest predictive performance (AUC-ROC: 0.84, 95% CI: 0.83–0.85; Accuracy: 86.0%; Recall: 85.0%; F1-score: 83.0%). SHAP analysis identified marital status, parity, maternal age, education, and healthcare-access barriers as key predictors. Association rule mining revealed critical multi-variable risk patterns; notably, being married combined with severe financial barriers to healthcare yielded a support of 18.02%, confidence of 48.34%, and a lift of 1.37 (a 37% increased risk above baseline). Conclusions Postpartum unmet need is driven by intersecting structural, economic, and systemic access barriers. Combining machine learning, SHAP, and association rule mining provides robust predictive power alongside interpretable, actionable insights to guide targeted reproductive health policy in low-resource settings.