A Comparative Analysis of Supervised Learning Methods in Maternal Health Risk Detection

Abstract

Maternal health risk involves conditions threatening a pregnant woman's well-being, requiring timely assessment to prevent complications and ensure safer pregnancy outcomes and Maternal health risk assessment is essential for early intervention. Machine learning offers accurate prediction methods, improving outcomes for pregnant women, especially in resource-limited settings. This study aims to evaluate and compare the performance of selected machine learning methods in predicting maternal health risk levels during pregnancy. Data were collected from various hospitals, community clinics, and maternal healthcare centers located in rural areas of Bangladesh using an IoT-based risk monitoring system. The dataset included six input features: Age, Systolic Blood Pressure, Diastolic Blood Pressure, Blood Sugar, Body Temperature, and Heart Rate. The outcome variable, Risk Level, was categorized into three classes: low, mild, and high. Three machine learning algorithms Random Committee (RC), Randomizable Filtered Classifier (RFC), and Nearest Neighbor with Generalization (NNge) were applied to the dataset to assess their classification accuracy. The primary objective was to identify the most efficient model for early detection and categorization of maternal health risks, potentially guiding healthcare providers in timely intervention. Overall, Random Committee (RC) emerges as the most effective algorithm in this study, offering the highest accuracy (85.21%), balanced sensitivity and specificity, and the greatest number of correctly classified instances. While RFC performs similarly and can be considered a reliable alternative, NNge, despite its strong specificity in the Low-risk class, demonstrates slightly lower accuracy and higher misclassification, making it less suitable for high-stakes medical decision-making where correctly identifying high-risk patients is critical.