Application of machine learning algorithms to model predictors of informed contraceptive choice among reproductive age women in six high fertility rate sub Sahara Africa countries.

INTRODUCTION

Informed contraceptive choice is declared when a woman selects a methods of contraceptive after receiving comprehensive information on available alternatives, side effects, and management if adverse effect happens. Access to contraceptive information is a fundamental right, crucial for reducing fertility and unintended pregnancies and related complications. Despite efforts to reduce fertility, Sub-Saharan Africa region is still accounts for over half of the global births due to low contraceptive use, high discontinuation rate, and unmet needs, often linked to uninformed contraceptive choice. While studies on informed contraceptive choice are available using classical regression analysis, the diverse nature of factors have not been systematically analyzed using machine learning algorithms. Hence, this study aimed to apply machine learning algorithms to model predictors of informed contraceptive choices among reproductive age women in six high fertility rate Sub Sahara Africa countries.

METHODS

This study used 11,706 weighted women aggregated from 6 high fertility rate countries in Sub Saharan Africa including Mali, Angola, Burundi, Nigeria, Gambia, and Burkina Faso, collected using stratified sampling techniques. Data cleaning, weighting, and descriptive statistical analyses were conducted using STATA version 17 and Excel 2019, while machine learning analysis was performed using Python 3.12. Furthermore, Random Forest, eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LGBM), Naïve Bayes, Decision Tree, Logistic Regression, and Adaptive Boosting (AdaBoost) were employed to predict informed contraceptive choice and to identify its predictors. Shapley Additive Explanations (SHAP) was used to assess the link between predictors and informed contraceptive choice. Accuracy and area under the curve (AUC), along with precision, recall, and F1 score, were used to evaluate the performance of the predictive models.

RESULTS

About 58% women receive informed choice of contraceptive methods, ranges 29% in Burundi to 77% in Burkina Faso. Moreover, the highest spatial clustering of informed choice of contraceptive methods cases was observed in Burkina Faso while the lowest is clustering was found in Angola. LGBM model achieved an accuracy of 73%, area under the curve (AUC) of 0.80, precision of 71, and recall of 77. The SHAP analysis revealed that health facility visits within 12 months, religion, source of contraceptive, exposure to family planning message, mobile ownership, education, wealth index, under five children, residence, and total life time partner were the top ten predictors of informed contraceptive choice.

CONCLUSION

Nearly six out of ten women received informed contraceptive choice, the magnitude is highest in Burkina Faso and lowest in Mali. Moreover, the highest spatial clustering of informed choice of contraceptive was observed in Burkina Faso while the lowest clustering was found in Angola. The LGBM classifier outperformed among machine learning algorithms and achieved 73% accuracy and an AUC of 0.80. Key factors influencing informed contraceptive choice were health facility visits, religion, contraceptive source, family planning messages, mobile ownership, education, wealth, residence, and lifetime partners. To enhance informed contraceptive choice, governments and policymakers should strengthen family planning education, expand healthcare services, and ensure equitable access to contraceptive information. Digital health solutions, especially mobile-based platforms, can also bridge information gaps. Integrating counseling into routine healthcare, training providers, and expanding mass media campaigns can enhance awareness. Engaging communities can help overcome social and religious barriers. Continuous monitoring and data-driven policy adjustments are essential for responsive interventions that address the evolving reproductive health needs in sub-Saharan Africa. Finally, we recommend that future research validate these findings using external data sources.