Machine learning models for predicting postpartum convulsions using clinical indicators from PMA Ethiopia data

Abstract

Postpartum convulsions, defined as seizure episodes occurring after childbirth during the postpartum period, remain a major cause of maternal morbidity and mortality in Ethiopia. This study aimed to develop and validate machine learning models to predict postpartum convulsions using clinical indicators from performance monitoring for action (PMA) Ethiopia data set. Eight machine learning algorithms were developed and evaluated. Recursive Feature Elimination (RFE) approach was used for feature selection, and class imbalance was addressed using up-sampling techniques. The data were preprocessed and split into training (70%) and testing (30%) sets. Model performance was assessed using accuracy, sensitivity, specificity, and the area under the receiver operating characteristic curve (AUC). Model interpretability was examined using Shapley additive explanations (SHAP). The neural network model achieved the best predictive performance (AUC = 0.868), followed by support vector machine (AUC = 0.833) and Random forest (AUC = 0.822). SHAP analysis identified migraine, convulsion, and abdominal pain during pregnancy as the most influential predictors, followed by timing of ANC visits, blood pressure, and urine testing, and lack of treatment for pregnancy-related complications. Machine learning models, particularly Neural Networks, showed promising performance in predicting postpartum convulsions using routinely collected clinical data. The use of explainable artificial intelligence may help identify women at higher risk and support timely clinical decision-making, potentially contributing to reduced maternal complications in low-resource settings.