Early Trimester Multi-Risk Prediction of Pregnancy Complications Using Explainable Ensemble Machine Learning

Abstract

Pregnancy complications such as gestational diabetes, hypertension, anemia, and preeclampsia increase health risks for both mothers and their unborn children. Health professionals need to identify these risks during the first trimester because this period is the most critical timeframe for effective preventive measures. Current risk-assessment methods predict only one specific risk instead of providing complete evaluations of early-stage health conditions. This study aims to develop an explainable ensemble machine learning framework for early trimester multi-risk prediction of pregnancy complications. An ensemble of Random Forest, XGBoost, and Support Vector Machine algorithms was used to develop a predictive model. The model was trained using maternal health data gathered during the first few weeks of pregnancy. To improve interpretability, model-agnostic explainability methods such as SHAP and LIME were used to provide both global and local insights into the model's predictions. The proposed ensemble model achieved strong predictive performance, with an accuracy of 88%, a macro F1-score of 0.89, and a macro ROC-AUC of 0.967. The explainability analysis identified key clinical features influencing pregnancy risk, improving transparency and supporting clinical understanding. The proposed framework demonstrates the effectiveness of combining ensemble learning with explainable AI for early pregnancy risk assessment. It provides accurate and interpretable predictions, enabling healthcare professionals to make informed decisions and improve maternal healthcare outcomes.