Ensemble-Based Model for Predicting Maternal Health Risk in Southwest Nigeria

Maternal mortality remains a critical public health concern, particularly in Nigeria, which continues to report one of the highest maternal death rates globally. This study proposes an innovative approach for predicting maternal health risks by integrating primary clinical data with a rule-based classification system and ensemble machine learning techniques. A dataset of 148 records was obtained from Ondo State University Teaching Hospital, encompassing key maternal health indicators. Given the lack of predefined class labels, a rule-based labeling framework adapted from a publicly available Kaggle dataset was applied. The data underwent preprocessing, including imputation for missing values and balancing using the Synthetic Minority Over-sampling Technique (SMOTE). Three ensemble machine learning models Voting, Stacking, and Bagging were developed and evaluated based on accuracy, precision, recall, and F1-score. Results showed that SMOTE markedly enhanced classification performance, with the Stacking ensemble achieving the highest accuracy (94.6%) and precision (97.1%). These outcomes highlight the potential of machine learning to enable early detection of maternal health risks and support improved decision-making in clinical settings.