# Insights into long-acting reversible contraceptive practices in Sub-Saharan Africa: A machine learning perspective

## TL;DR

This study uses machine learning to identify factors influencing contraceptive use in Sub-Saharan Africa and suggests targeted interventions to improve access.

## Contribution

The novel use of machine learning to analyze and predict LARC adoption in Sub-Saharan Africa, identifying actionable policy recommendations.

## Key findings

- Random Forest achieved excellent performance (AUC-ROC: 1.00) in predicting LARC use.
- Household size, age at first contraceptive use, and current age were the strongest predictors of LARC adoption.
- Socio-cultural factors like religion and marital status had negligible impact on LARC use.

## Abstract

Long-acting reversible contraceptives (LARCs) are critical for reducing maternal mortality and unintended pregnancies, yet adoption remains low in Sub-Saharan Africa (SSA) due to systemic inequities, cultural barriers, and fragmented healthcare access. Despite global advancements, only 8% of women in SSA use LARCs, underscoring the need for data-driven insights to address this gap. This study applies machine learning (ML) to identify key predictors of LARC use and guide interventions.

Nationally representative data from 14,275 women across nine SSA countries were analyzed. Preprocessing included k-NN imputation and advanced class balancing (SMOTEENN). Feature engineering derived interaction terms (age×household size, education×media exposure) with SHAP-driven selection. Eight ML models were trained and hyperparameter-tuned using stratified cross-validation.

After hyperparameter tuning and class balancing, Random Forest achieved excellent discriminative performance (AUC-ROC: 1.00). Key predictors were household size (SHAP = 0.464), age at first contraceptive use (0.396), and current age (0.376). Socio-cultural factors (religion, marital status) showed negligible impact and were excluded. LARC uptake remained critically low (3.3%) with persistent rural-urban disparities.

The model’s key predictors directly inform policy; we recommend: 1) Mobile clinics for young women in large households, targeting the two strongest negative predictors (young age and large household size), 2) Media campaigns tailored to educated populations, leveraging the significant interaction between education and media exposure, and 3) Adolescent-focused education on contraceptive timing, addressing the critical predictor of age at first use.