Clinical And Laboratory Determinants Of Preeclampsia Onset Using Ensemble Machine Learning Prediction And Comparison Among Stacking Ensemble, Neural Network With And Without Stacking Ensemble Approach

Preeclampsia is a pregnancy-related hypertensive illness that bears increased maternal and fetal morbidity and mortality. For early risk stratification of preeclampsia, the development of models that can forecast its onset is crucial. This work analyzes maternal clinical and laboratory features using bagging- and boosting-based ensemble models in comparison with a neural network stacking framework that incorporates these models. Experimental results show that Random Forest yields the highest performance of 81% for late-onset preeclampsia and maintains robust performance for early-onset preeclampsia. In contrast, the boosting models XGBoost and CatBoost yield competitive performance with an accuracy of 77% and 71%, respectively. However, their performance is still inferior to that from Random Forest. Hematological parameters, especially total leukocyte count and platelet count, are major predictors influencing model performance and misclassification patterns across all the ensemble models. On the other hand, neural network-based stacking ensemble models yielded more balanced and reliable predictions, with great value for clinical risk assessment. The developed models can enhance the early risk stratification of preeclampsia and support decisions for obstetric care.