Improving maternal and neonatal health outcomes in Kenya by leveraging machine learning for the timely detection of preeclampsia

Abstract

Preeclampsia remains a leading cause of maternal and neonatal morbidity and mortality globally, with its burden disproportionately high in low-resource settings such as Kenya. Timely diagnosis is critical to improving outcomes, yet healthcare systems often lack the tools for early and reliable risk assessment. This study aimed to develop an AI-powered machine learning (ML) model to predict preeclampsia among pregnant women in Kenya, based on routinely collected clinical and demographic data. The specific objectives were: (a) to identify key risk factors using statistical analysis and clinical insight; (b) to develop and evaluate various ML models for preeclampsia prediction; (c) to determine the most accurate model for risk classification; and (d) to deploy the predictive system across multiple platforms for practical use. Using a dataset of 2,925 records collected from hospitals in Kenya's coastal region, the study applied extensive preprocessing to ensure data quality and clinical relevance. Variables included age, pre-pregnancy weight, systolic and diastolic blood pressure, proteinuria levels, parity, and history of hypertensive disorders. Five individual ML models—Logistic Regression, Random Forest, XGBoost, Support Vector Machine (SVM), and Decision Tree—were trained and evaluated. The XGBoost model emerged as the best individual performer, achieving an accuracy of 98.46%, precision of 95.59%, recall of 97.74%, and F1-score of 96.65%. However, a hybrid stacking classifier, which combined the predictions of multiple base models with Logistic Regression as a meta-learner, outperformed all individual models. It achieved an accuracy of 98.12%, a recall of 99.25%, and an F1-score of 96.00%, making it the most reliable for clinical deployment. The final model was deployed as a functional web-based application, allowing healthcare providers to input patient data and receive immediate risk assessments. This implementation underscores the potential of AI in enhancing prenatal care by enabling early intervention. The study concludes that ensemble-based ML models, especially stacking classifiers, provide a robust and scalable solution for preeclampsia risk prediction. Future work should focus on expanding the dataset to include diverse populations, integrating additional biomarkers, and developing mobile and EHR-compatible interfaces for broader reach in underserved areas.