Machine Learning-Based Maternal Health Risk Assessment: A Comparative Analysis of Classification Algorithms for Predicting Risk Levels During Pregnancy

Background: Maternal health risk assessment remains a critical challenge in healthcare, particularly in resource-limited settings where early identification of high-risk pregnancies can significantly impact maternal and fetal outcomes. This study evaluates the performance of multiple machine learning algorithms for predicting maternal health risk levels using physiological parameters.

Methods: We analyzed a dataset of 1014 pregnant women from Kaggle, incorporating six key features: age, systolic blood pressure, diastolic blood pressure, blood sugar levels, body temperature, and heart rate. Risk levels were classified as mild (0), moderate (1), and severe (2). Four machine learning algorithms were implemented and compared: Logistic Regression, Random Forest, Support Vector Machine (SVM), and K-Nearest Neighbors (KNN).

Results: Random Forest and SVM achieved perfect classification performance with 100% accuracy, precision, recall, and F1-scores across all risk categories. Logistic Regression demonstrated strong performance with 98% overall accuracy, showing minor challenges in recall for moderate risk cases (93%). KNN achieved 98% accuracy with balanced performance across risk categories, though slightly lower precision for mild risk cases (95%).

Conclusion: Machine learning algorithms, including Random Forest and SVM, show promise in predicting maternal health risks; however, further validation across diverse populations is essential before clinical adoption.