Early Prediction of Maternal Health Risks Using Machine Learning Techniques

Abstract

Accurate prediction of maternal health risks is critical for enabling early interventions to reduce pregnancy-related complications. This study evaluates six machine learning classifiers XGBoost, LightGBM, CatBoost, Random Forest, Gradient Boosting, and k-Nearest Neighbors (KNN) to classify risk levels using clinical parameters such as age, systolic diastolic blood pressure, blood glucose, heart rate, and body temperature. A rigorous feature importance analysis via Random Forest identified age and blood glucose levels as the most influential predictors, optimizing model efficiency. To ensure reliability, repeated stratified k-fold cross-validation was employed, minimizing bias and enhancing generalizability.

Among all classifiers, XGBoost achieved the highest accuracy 0.97%, demonstrating superior performance in risk stratification through its regularization and ensemble learning framework. In contrast, KNN recorded the lowest accuracy 0.94%, yet maintained clinical relevance due to its simplicity and interpretability. LightGBM, CatBoost, Random Forest, and Gradient Boosting also contributed robust results, further validating the efficacy of ensemble methods. The findings underscore the significance of feature selection, with age and blood glucose emerging as pivotal determinants of maternal risk. This study provides a scalable, data-driven framework for healthcare systems to prioritize high-risk pregnancies, offering actionable insights for timely interventions and informed clinical decision-making. By integrating these models into maternal care protocols, practitioners can enhance early diagnosis and reduce preventable morbidity, advancing equitable healthcare outcomes globally.