A Comparative Analysis and Interpretation of Random Forest, XGBoost, and SVM (RBF) Models with Explainable AI for Pregnancy Risk Classification

Abstract

Accurate and early identification of high-risk pregnancies is critical for reducing maternal and neonatal mortality, yet conventional screening methods often lack the precision to capture complex, interacting risk factors. While machine learning (ML) offers a powerful solution, its "black box" nature hinders clinical trust and adoption. This study develops and compares three machine learning models Random Forest, XGBoost, and Support Vector Machine (SVM) for binary pregnancy risk classification using a comprehensive set of maternal clinical data. To move beyond a purely predictive framework, we employed a state-of-the-art Explainable AI (XAI) technique, SHAP (SHapley Additive exPlanations), to interpret the decision-making process of the best-performing model. The Random Forest model demonstrated superior performance, achieving an accuracy of 95.2% and a sensitivity of 92.7% for identifying the high-risk class on the test set. The SHAP analysis successfully demystified the model, revealing that fetal heart rate (djj), systolic blood pressure (td_sistolik), maternal age (usia), and hemoglobin (hb) level were the most influential predictors. The interpretation also confirmed that the model learned clinically relevant relationships, such as low hemoglobin and high blood pressure contributing to increased risk. Our findings demonstrate that combining high-performance machine learning with XAI provides a robust framework for developing not just accurate, but also transparent and clinically trustworthy, tools. This approach offers a significant step towards enhancing data-driven decision support in antenatal care.