Ensemble-Based Maternal Health Risk Prediction using Clinical Physiological Data

Maternal health complications remain a significant public health concern in Nigeria, particularly in resource-constrained healthcare settings where access to advanced diagnostic technologies is limited. Early identification of women at increased risk can support timely intervention and improve maternal health outcomes. This study presents a machine learning approach for maternal health risk prediction using routinely collected physiological parameters from148 maternal health records obtained from a public health facility in Ondo State, Nigeria. The dataset comprises age, body temperature, weight, height, systolic blood pressure, and diastolic blood pressure. A clinically informed rule-based approach was used to derive maternal health risk labels from the observed physiological parameters because the clinical records did not contain predefined risk categories. The data were preprocessed to address missing values and prepare the physiological features for model development. The dataset was then stratified into 70% training and 30% testing sets, and SMOTE was applied to the training data to address class imbalance. Voting, Stacking, and Bagging ensemble learning approaches were subsequently developed for maternal health risk prediction. Performance evaluation showed that the Stacking approach outperformed the other ensemble models with accuracy (94.6%) and precision (97.1%) on the held-out testing data. The results presented in Table 1 shows that stacking ensemble achieved the best overall performance, with 94.60% accuracy, 97.10% precision, and 94.30% F1-score. Despite the limited dataset size, the findings show the potential of ensemble-based learning to support the early identification of women with increased maternal health risk and contributes to the broader objective of the Sustainable Development Goal 3, Target 3.1 (SDG 3.1).