Machine learning for contraceptive use prediction
1
School of Information Science, Addis Ababa University, Addis Ababa University, Ethiopia
2
School of Public Health, Addis Ababa University, Addis Ababa, Ethiopia
Popul. Med. 2026;8(Supplement Supplement 1):A944
ABSTRACT
INTRODUCTION:
Modern contraceptive use is a cornerstone of reproductive health 1,2, contributing significantly to reducing healthcare expenditures, maternal mortality, and unsafe abortions, and enhanced socio-economic outcomes for women 1,3–9. However, coverage remains uneven, and the ability to accurately identify women most likely to use or not use contraception is limited. Machine learning (ML) offers new opportunities to improve predictive performance and inform targeted family planning interventions. This study applied a comprehensive, multi metric evaluation framework to identify the most accurate and reliable model for predicting modern contraceptive use among Ethiopian women.
METHODS:
We analyzed data from the PMA Ethiopia surveys (2019–2023). Eight ML algorithms: Logistic Regression, Gaussian Naïve Bayes, Support Vector Classifier, Multilayer Perceptron(MLP), Decision Tree, Random Forest (RF), AdaBoost, and Gradient Boosting Machine were trained using a standardized pipeline with weighted preprocessing, 10 fold stratified cross validation, Bayesian hyperparameter optimization, and Platt probability calibration. Models were trained on pooled 2019–2021 data and externally validated on the temporally held out 2023 dataset. Performance was assessed using discrimination, calibration, and threshold dependent classification metrics with optimized thresholds selected to maximize F1 for the positive class.
RESULTS:
Across models, discrimination was high (ROC–AUC range: 0.821–0.866). RF achieved the best overall performance based on the predefined multi metric criteria: highest PR–AUC (0.655), lowest Brier score (0.133), and one of the highest F1 scores after threshold optimization (F1=0.637 at threshold 0.30). Threshold tuning substantially improved recall for modern contraceptive users across all models. The MLP demonstrated strong discrimination but inferior calibration relative to Random Forest.
CONCLUSIONS:
Random Forest provided the most reliable and well calibrated predictions and demonstrated strong generalization to the 2023 survey. ML, when paired with rigorous model evaluation and calibration, can enhance the identification of groups with low modern contraceptive use and support tailored, data driven family planning strategies in Ethiopia.
Modern contraceptive use is a cornerstone of reproductive health 1,2, contributing significantly to reducing healthcare expenditures, maternal mortality, and unsafe abortions, and enhanced socio-economic outcomes for women 1,3–9. However, coverage remains uneven, and the ability to accurately identify women most likely to use or not use contraception is limited. Machine learning (ML) offers new opportunities to improve predictive performance and inform targeted family planning interventions. This study applied a comprehensive, multi metric evaluation framework to identify the most accurate and reliable model for predicting modern contraceptive use among Ethiopian women.
METHODS:
We analyzed data from the PMA Ethiopia surveys (2019–2023). Eight ML algorithms: Logistic Regression, Gaussian Naïve Bayes, Support Vector Classifier, Multilayer Perceptron(MLP), Decision Tree, Random Forest (RF), AdaBoost, and Gradient Boosting Machine were trained using a standardized pipeline with weighted preprocessing, 10 fold stratified cross validation, Bayesian hyperparameter optimization, and Platt probability calibration. Models were trained on pooled 2019–2021 data and externally validated on the temporally held out 2023 dataset. Performance was assessed using discrimination, calibration, and threshold dependent classification metrics with optimized thresholds selected to maximize F1 for the positive class.
RESULTS:
Across models, discrimination was high (ROC–AUC range: 0.821–0.866). RF achieved the best overall performance based on the predefined multi metric criteria: highest PR–AUC (0.655), lowest Brier score (0.133), and one of the highest F1 scores after threshold optimization (F1=0.637 at threshold 0.30). Threshold tuning substantially improved recall for modern contraceptive users across all models. The MLP demonstrated strong discrimination but inferior calibration relative to Random Forest.
CONCLUSIONS:
Random Forest provided the most reliable and well calibrated predictions and demonstrated strong generalization to the 2023 survey. ML, when paired with rigorous model evaluation and calibration, can enhance the identification of groups with low modern contraceptive use and support tailored, data driven family planning strategies in Ethiopia.
| eISSN: | 2654-1459 |
We process personal data collected when visiting the website. The function of obtaining information about users and their behavior is carried out by voluntarily entered information in forms and saving cookies in end devices. Data, including cookies, are used to provide services, improve the user experience and to analyze the traffic in accordance with the Privacy policy. Data are also collected and processed by Google Analytics tool (more).
You can change cookies settings in your browser. Restricted use of cookies in the browser configuration may affect some functionalities of the website.
You can change cookies settings in your browser. Restricted use of cookies in the browser configuration may affect some functionalities of the website.
