Health Risk Detection and Early Screening of Maternal Health in Pregnant Women Using a Stacking Machine Learning Models

The increasing prevalence of health risks among pregnant women, requires early and accurate detection systems to improve maternal and fetal outcomes. This study proposes a machine learning-based early detection system for health risks such as diabetes, hypertension, and maternal health dataset. The model implements a stacked ensemble meta-model combining XGBoost and Random Forest as base classifiers, and Gradient Boosting as the meta-classifier. Three datasets were used: Maternal Health, Diabetes, and Hypertension Prediction datasets. Data preprocessing included cleaning, standardizing feature names, and balancing class distribution using the SMOTE technique. Hyperparameter tuning was performed using GridSearchCV to optimize model performance. The model was evaluated using accuracy, precision, recall, F1-score, and confusion matrix metrics. The results demonstrated that the proposed meta-model achieved accuracies of 86% for the Maternal Health dataset, 82% for the Diabetes dataset, and 92% for the Hypertension dataset. These findings indicate that the stacked meta-model delivers reliable and consistent performance in predicting health risks in pregnant women. This system offers potential as an early screening tool, particularly in rural areas with limited healthcare access, to support timely and targeted maternal healthcare services.