Tuesday July 22, 2025
FILE - A Somali mother holds her child outside their home in a rural village, where limited access to health services and education continues to shape fertility decisions. A new study highlights age, childbirth history, and access to care as key factors influencing women’s reproductive preferences in Somalia.
Mogadishu (HOL) — A Somali-led study has used machine learning to identify age, childbirth history, and access to health care as the strongest predictors of whether women want more children, insights that could reshape family planning efforts in one of the world’s highest-fertility countries.
Published this month in Scientific Reports, the study analyzed data from nearly 9,000 Somali women of reproductive age and applied machine learning tools to predict fertility preferences with 81% accuracy. The study was authored by Jamilu Sani, Salad Halane, Abdiwali Mohamed Ahmed, and Mohamed Mustaf Ahmed. The research represents the first of its kind in Somalia and offers actionable insights for designing reproductive health programs in one of the world’s highest fertility environments.
The researchers used a Random Forest model, a form of ensemble machine learning, to identify the most influential predictors, followed by SHAP (Shapley Additive Explanations) values to interpret the model’s decision-making process at both the population and individual level.
“The model indicates that age is the most influential factor in determining a woman's desire for more children, with older women being more likely to express a preference for no more children,” the authors wrote.
Somalia has one of the world’s highest fertility rates—estimated at 6.4 children per woman—with persistent challenges in access to contraception, maternal health, and education, particularly in rural and nomadic communities. The Somali government has prioritized reproductive health in its national development plans, but remains constrained by data gaps and resource limitations.
This study, based on the 2020 Somalia Demographic and Health Survey (SDHS), offers granular insights across all 18 administrative regions and residential zones (urban, rural, and nomadic). The findings could help shape evidence-based interventions to improve maternal outcomes and expand access to family planning services.
Key findings
Among the seven machine learning algorithms tested, the Random Forest classifier yielded the best performance with an accuracy of 81%, a precision of 78%, a recall of 85%, and an F1-score of 81%.
The five most influential features ranked by SHAP values were:
- Age
- Region
- Time since last birth
- Distance to health facility
- Number of living children
“Women aged 45–49 were 5.32 times more likely to prefer no more children compared to those aged 15–19 (OR = 5.32; 95% CI: 3.67, 7.73),” the report states.
Similarly, women with seven or more children were over three times as likely to want no more children as those with none (OR = 3.16; 95% CI: 1.79, 5.59).
Access to health care played a major role:
- 62.5% of respondents said distance to a health facility was a “big problem.”
- Women who reported that distance was “not a big problem” were 1.54 times more likely to desire more children (OR = 1.54; 95% CI: 1.28, 1.84).
Education was extremely limited:
- 83.1% of participants had no formal education.
- Only 1.4% had higher education.
Regional disparities were substantial:
- Women in Lower Juba were 2.8 times more likely to desire no more children than those in Awdal.
- Galgaduud showed one of the lowest likelihoods of preferring to stop childbearing (OR = 0.63; 95% CI: 0.41, 0.97).
Employment appeared to have a weaker correlation:
- Only 7.95% of women reported working in the 12 months preceding the survey.
To make the results transparent, the study employed SHAP values to identify the contribution of each feature to the model’s predictions.
“This granularity helps in identifying specific factors influencing individual decisions, which is crucial for designing personalized interventions and understanding regional or demographic variations,” the authors note.
This interpretability is especially important in low-resource settings like Somalia, where national health strategies require precise targeting.
The study recommends using these findings to:
- Improve outreach in underserved regions
- Expand mobile health services
- Promote female education as a long-term strategy
“The insights gained from SHAP values can be instrumental in shaping targeted health interventions and policies,” the authors wrote. They also call for more “contextual and longitudinal studies to further understand the causality and dynamics” of fertility preferences.
The authors emphasize that machine learning models like Random Forest not only outperform traditional statistical methods in prediction, but also provide “robust and interpretable tools for health policymakers.”