Improving Maternal Risk Analysis in Public Health Systems

There are several efforts in intelligent approaches in the literature dealing with maternal death risk prediction. Solutions focused on surveillance and monitoring maternal health contributes to reduce mortality rates, especially in low and middle-income countries. Data required by artificial intelligence systems are usually sensitive and restricted by privacy policies. High quality and trusted maternal health data are essential to obtain reliable predictive models. This study applies the Recursive Feature Elimination (RFE) strategy associated with decision tree-based classifier identifying a relevant set of features among an extensive list of maternal predictive information considered in a decision-making process. This study applies a systematic process of data preparation, analysis, and modeling to develop trusted models from maternal Electronic health registries (eRegistries). Also, this research presents an experiment pipeline to evaluate six well-known supervised machine learning models, namely Random Forest (RF), Support Vector Machine (SVM), Multilayer Perceptron (MLP), Adaptive Boosting (AdaBoost), Decision Tree (DT), and Gaussian Naive Bayes (GNB), with different combinations of ranked features. Results show that the feature ranking strategy was useful to reduce data dimensionality without affecting the performance of predictive models. The RFE-based predictive models achieves high Accuracy (ACC) and Area Under the Receiver Operating Characteristic Curve (AUC) with only eight maternal features.