Maternal Health Risk Assessment with Interpretable Machine Learning: Evidence from Bangladesh

Maternal health remains a critical public health concern, particularly in regions with limited healthcare resources. This study adopts a data-driven approach to predict maternal risk levels during pregnancy. Using real-world clinical data from a general hospital in Kurigram, Bangladesh-including age, blood pressure, blood sugar, body mass index (BMI), mental health indicators, and medical history-we developed predictive models to classify pregnancies as binary classification (high risk or low risk). Eight machine learning algorithms were trained and evaluated: Random Forest, Support Vector Machine, Neural Network, XGBoost, Logistic Regression, K-Nearest Neighbors, Gaussian Naive Bayes, and Gradient Descent. Preprocessing included label encoding, min-max scaling, and feature selection through chi-square tests and ANOVA. To enhance interpretability, SHAP and LIME were applied to identify feature contributions. Among all models, Random Forest achieved the highest performance with an accuracy of 97%. Results highlight BMI, blood sugar, and preexisting diabetes as key predictors. The findings demonstrate how combining statistical analysis, machine learning, and explainable AI can support healthcare professionals in identifying high-risk pregnancies early and making evidence-based decisions for safer maternal care.