Machine Learning Algorithm for Maternal Health Risk Classification with SMOTE and Explainable AI

Targeted obstetric care refers to being specific with the caretaking of a pregnant woman based on her symptoms and risk levels. It is a crucial task because pregnancy-related complications can be life-threatening. Pregnant women's body undergo a variety of changes, which may be mental, emotional, or physical. It is possible to analyze these changes to assess the health complications. These include blood pressure and blood sugar readings, which can be examined to determine a pregnant woman's health risk level. Machine Learning (ML) algorithms are employed to determine maternity health risks based on a woman’s health statistics. In this work, several ML algorithms, have been applied to the maternal health risk dataset to predict the mother’s risk level. Moreover, parameterized tuning has been applied to improve the performances of these conventional algorithms. The performance of these algorithms is studied before and after applying the Synthetic Minority Over-sampling Technique (SMOTE) technique to balance the data. The results exhibit that after SMOTE enhancement, the XGBoost upon hyperparameter tuning gives an accuracy of 88.89% in predicting mothers’ risk level. Hence, XGBoost can be deployed in the field of providing targeted obstetric care by accurately predicting health risk levels. This work also embeds the Explainable AI (XAI) using SHapley Additive exPlanations (SHAP) and LIME (Local Interpretable Model-agnostic Explanations) to provide trust in the ML model to gynecologists. Various plots are generated using LIME, SHAP, and ELI5 to showcase the ML model’s interpretability on global and local levels.