Abstract
High-risk pregnancies are a major contributor to maternal and neonatal morbidity and mortality, particularly in resource-constrained healthcare systems such as Pakistan, where delayed diagnosis and limited clinical decision-support tools hinder timely intervention. This study developed and evaluated an integrated machine learning-based risk prediction model for the early detection of high-risk pregnancies using clinical, demographic, obstetric, and socio-economic data. A quantitative cross-sectional design was employed, and data from 600 pregnant women were analyzed using multiple machine learning algorithms, including Logistic Regression, Support Vector Machine, Random Forest, and XGBoost. Model performance was assessed using accuracy, precision, recall, F1-score, and AUC-ROC. Results indicated that XGBoost achieved the highest predictive performance (accuracy = 91.8%, AUC-ROC = 0.94), outperforming other models. Key risk factors identified included blood pressure, hemoglobin levels, antenatal care visits, obstetric history, and socio-economic status. To enhance interpretability and clinical trust, Explainable Artificial Intelligence (SHAP) was applied, revealing transparent contributions of individual predictors. The findings demonstrate that integrated AI-based predictive models significantly improve early risk detection and clinical decision-making in maternal healthcare. The study concludes that such systems can support timely interventions and improve maternal and neonatal outcomes in resource-limited settings.