Peer-reviewed article
Explainable machine learning for predicting preterm birth and associated factors in Sub Saharan Africa
Makda Fekadie Tewelgne, Tirualem Zeleke Yehuala, Mekuriaw Nibret Aweke, Habtamu Wagnew Abuhay, Miteku Andualem Limenih et al. ·
Summary
Written by AI, not yet AICS's view · claude-sonnet-5This study used machine learning to predict preterm birth risk using health survey data from eleven Sub-Saharan African countries, covering over 400,000 women. The Random Forest model was the most accurate at predicting preterm birth. The researchers identified factors linked to higher predicted risk, including low maternal education, being unmarried, low household wealth, and insufficient antenatal care visits. The study explains which factors matter most to the prediction model but does not prove these factors cause preterm birth.
Key points
- Random Forest achieved 90.20% accuracy and 96.92% ROC-AUC, the best of four algorithms tested.
- Key predictors included low maternal education, unmarried status, lack of media exposure, lower wealth, and maternal age 35+.
- Other predictors were substance use, rural residence, employment, anemia, abnormal BMI, fewer than 4 antenatal visits, and short birth interval.
- SMOTE was used to handle class imbalance and improved identification of preterm birth cases.
- Findings represent predictive associations, not causal relationships.
Limitations
The study is based on cross-sectional survey data and explicitly states its findings are predictive associations rather than causal relationships, with caution needed when generalizing across different country contexts.
Abstract
Preterm birth remains a leading contributor to neonatal mortality and long-term morbidity worldwide, with Sub-Saharan Africa bearing a disproportionate share of the burden. Structural inequities, limited maternal healthcare access, and socioeconomic vulnerabilities are associated with higher model-predicted probability of preterm birth. Despite its significance, large-scale predictive modeling of preterm birth in this region remains limited. This study aimed to develop robust machine-learning–based prediction models and identify key factors associated with preterm birth using recent multi-country Demographic and Health Survey (DHS) data from (2019–2024). A pooled cross-sectional analysis was conducted using nationally representative DHS data from eleven Sub-Saharan African countries, including 411,863 women aged 15–49 years. Four supervised machine learning algorithms were trained and tested using an 80/20 train–test split. Model performance was evaluated using accuracy, precision, recall, F1-score, and ROC-AUC. Model interpretability was assessed using SHapley Additive exPlanations (SHAP). The Random Forest model demonstrated the highest predictive performance for preterm birth, achieving an accuracy of 90.20% and ROC-AUC of 96.92%, outperforming the Decision Tree (accuracy of 89.2% and ROC-AUC of 92.19%), and other classifiers. The use of SMOTE for class imbalance handling and evaluation with precision–recall metrics improved model robustness in identifying minority class outcomes. SHAP-based feature importance analysis identified key predictors of the outcome,including low maternal educational attainment,unmarried marital status, lack of media exposure, lower household wealth index, and advanced maternal age (≥ 35 years). Additional important predictors were substance use,rural residence,employed women,anemia, abnormal maternal body mass index, insufficient antenatal care attendance (< 4 visits), and short birth interval. Overall, all these factors were associated with an increased predicted probability of preterm birth. Machine learning models, particularly ensemble methods, demonstrated strong predictive performance for preterm birth in Sub-Saharan Africa. Random Forest performed best, highlighting the effectiveness of ensemble learning in handling complex and imbalanced datasets. Key socio-demographic and maternal health factors were consistently associated with preterm birth in a predictive (non-causal) framework. The integration of explainable artificial intelligence enhances model transparency and supports interpretation of predictive patterns in maternal health data. Findings should be interpreted as predictive associations rather than causal relationships, and caution is required when generalizing across heterogeneous country contexts.