Machine learning predictive system to predict the risk of developing pre-eclampsia

OBJECTIVES: To develop a machine learning (ML)-based predictive model for assessing the risk of pre-eclampsia using routinely collected clinical data. METHODS: We retrospectively analysed data from 2444 pregnant women who delivered at Chi Mei Medical Center between 2015 and 2019, excluding patients under 20 years old. Pre-eclampsia was defined as blood pressure >140/90 mm Hg with proteinuria. Key features included gestational age, body weight, blood pressure and medical history. Five ML models were trained-logistic regression, random forest, light gradient boosting machine, extreme gradient boosting (XGBoost) and multilayer perceptron-using a 70/30 train-validation split. Synthetic minority oversampling technique was applied to address class imbalance. SHapley Additive exPlanations (SHAP) analysis was used for feature importance. RESULTS: Among the five models, XGBoost showed the best performance with the highest accuracy, sensitivity, specificity and an area under the receiver operating characteristics curve of 0.921. SHAP analysis identified diastolic blood pressure, systolic blood pressure and urine glucose as top predictors. DISCUSSION: Our findings demonstrate that ML, particularly XGBoost, can effectively predict the risk of pre-eclampsia using standard clinical data. This approach avoids the need for expensive tests while maintaining high accuracy. CONCLUSION: The XGBoost model offers a cost-effective and accurate method for pre-eclampsia risk prediction, enabling real-time assessment and supporting early intervention. Future studies will focus on larger data sets and clinical integration.