Explainable machine learning algorithms for predicting maternal obstetric complications in Ethiopia: Evidence from a retrospective cohort in southern Ethiopia

Introduction Early identification of high-risk pregnancies is crucial, yet conventional approaches often miss complex clinical and contextual factors. This study measured prevalence and determinants in Southern Ethiopia and applied explainable machine learning as a clinical decision-support tool, linking prediction to actionable digital health insights. Methods We conducted a retrospective cohort study of 3,954 mother–infant pairs using routine records from pregnancy to the postpartum. Five supervised machine learning algorithms - Logistic Regression, Random Forest, Support Vector Machine, an Artificial Neural Network and XGBoost were developed and validated. Performance was assessed using AUC, specificity, sensitivity, calibration, and F1-score. SHAP-based analysis enhanced interpretability, revealing the contribution of each predictor at both individual and cohort levels, supporting practical integration into digital maternal health systems. Results Obstetric complications occurred 16.3% of mothers, with higher incidence in rural settings. XGBoost achieved the highest predictive performance (AUC 0.86). Key predictors identified young maternal age, unplanned pregnancy, low education, previous complications, inadequate antenatal care, anemia, hypertension and long travel distance to health facilities. Conclusions Obstetric complications remain common in Southern Ethiopia. By applying explainable machine learning, this study not only predicts high-risk pregnancies with high accuracy but also provides actionable insights for clinical decision-making and digital health implementation.