Contraceptive ever-use among women of reproductive age in Northern Ghana: a comparative analysis of logistic regression, LASSO penalised regression, and random forest machine learning approaches.

Background: Contraceptive uptake in Northern Ghana remains low despite high awareness, suggesting barriers not fully captured by additive models. This study compared logistic regression, LASSO, and Random Forest to predict contraceptive ever-use and identify determinants among women in Tamale Metropolis. Methods: A facility-based cross-sectional survey (January-April 2023) included 414 women aged 15-49 from three public health facilities. Outcome: contraceptive ever-use. Twelve predictors included knowledge/attitude scores, wealth tertile, education, parity, marital status, occupation, age, cultural barriers, health worker attitudes, and information source. Model discrimination was assessed via AUC with 95% CIs; LASSO used 10-fold cross-validation ( Results: Random Forest achieved highest discrimination (AUC=0.865, 95% CI:0.83-0.90), outperforming logistic regression (AUC=0.727) and LASSO (AUC=0.723) (both Conclusions: Random Forest provided better discrimination and may complement regression for prediction, though regression remains essential for interpretable effect estimates. Attitude change, education, and counselling quality are priority targets. The inverse wealth association warrants qualitative investigation, as polygyny, religiosity, and partner dynamics were unmeasured.