Machine Learning Identification of High-Risk Profiles for Unintended Pregnancy in the Benin Republic: Evidence from DHS Data

Abstract

Unintended pregnancy remains a major reproductive health challenge in Benin Republic, with adverse consequences for women’s health and socio-economic wellbeing. This study aimed to identify and profile women at elevated risk of unintended pregnancy in Benin using machine learning techniques to support equity-driven interventions. We analyzed data from the 2017–2018 Benin Demographic and Health Survey for women aged 15–49 whose most recent birth was live. The outcome variable was unintended pregnancy, defined as mistimed or unwanted. We applied Synthetic Minority Oversampling Technique to address class imbalance and tested eight supervised learning algorithms. Random Forest emerged as the best-performing classifier and was used for variable importance ranking and high-risk profiling. Of 8,993 eligible women, 26% reported unintended pregnancy. Random Forest achieved 96% accuracy and 1.00 AUROC under 10-fold cross-validation. Key predictors included region, religion, education level, number of children, wealth index, and marital status. Top profiles revealed heightened risk among Christian women in Littoral with 6–7 children, no formal education, middle income, and cohabiting. Notably, risk transcended poverty, with some affluent women also affected. Targeting education, family planning access, and economic empowerment—especially in Littoral and Donga—will enhance reproductive autonomy and health equity.