MACHINE LEARNING MODELS FOR EARLY PREDICTION OF PREECLAMPSIA

Abstract

Preeclampsia continues to be a leading contributor to maternal and perinatal mortality worldwide, and its incompletely understood pathophysiology has long limited the accuracy of conventional risk-scoring tools based on maternal characteristics alone. Machine learning has been proposed as a way to capture the complex, non-linear interactions between demographic, biochemical and Doppler ultrasound variables that traditional regression-based models tend to oversimplify [1]. A systematic review evaluating eleven machine learning models published between 2020 and 2025 across the United States, South Korea, China, Romania, Mexico, Australia and Spain found considerable variability in predictive performance, with sample sizes ranging from a few hundred to nearly fifty thousand pregnancies, underscoring both the promise and the current fragmentation of this research field [2].