INTRODUCTION:
Contraceptive discontinuation remains a major barrier to achieving effective family planning outcomes in Ethiopia 1,2. This study applies machine learning techniques to predict discontinuation of modern contraceptive methods among women aged 15–49, aiming to generate actionable insights for programmatic decision making.

METHODS:
The analysis utilized the PMA Ethiopia dataset with weighted sample of 33301 women records. Eight machine learning algorithms: Adaptive Boosting (AdaBoost), Artificial Neural Network (ANN), eXtreme Gradient Boosting (XGBoost), k-Nearest Neighbor (KNN), Logistic Regression (LR), Naïve Bayes (NB), Random Forest (RF), and Support Vector Machine (SVM) were trained and evaluated. The workflow incorporated rigorous data preprocessing and cross validation to ensure robust model performance and minimize overfitting.

RESULTS:
The two-year contraceptive discontinuation rate was 29.65%. Among the tested models, the RF algorithm demonstrated superior predictive performance, achieving an accuracy of 85.84%. Key predictors of discontinuation included method type, household decision making dynamics, educational attainment, wealth status, pregnancy-related attitudes, source of contraceptive method, region of residence, and indicators of sexual and reproductive health empowerment.

CONCLUSIONS:
Findings underscore the complex interplay of socio economic, behavioral, and contextual factors influencing contraceptive discontinuation among Ethiopian women. The application of machine learning offers valuable predictive capabilities that can guide evidence based, targeted interventions. Strengthening family planning programs will require strategies tailored to the diverse needs and circumstances of women across Ethiopia.