Comparative Evaluation of Machine Learning Models for Maternal Health Risk Prediction Using IoT-Based Data

Abstract

Maternal health risk prediction is crucial for reducing maternal mortality, a key concern of the United Nations’ Sustainable Development Goals (SDGs). The main objective of the study is to evaluate the predictive performance of various machine learning (ML) models, identify the most significant physiological risk factors, and provide a data-driven approach for improving maternal healthcare interventions. This study utilizes ML models to classify maternal risk levels using an Internet of Things (IoT)-based dataset obtained from the UCI ML repository, which was collected from rural areas of Bangladesh. The dataset comprises 1014 observations with six physiological independent variables: Age, Systolic Blood Pressure, Diastolic Blood Pressure, Blood Sugar, Body Temperature, and Heart Rate. The categorical dependent variable represents the Risk Level. The Boruta algorithm and the Regularized Random Forest (RRF) method are applied for feature selection to enhance model efficiency. Various ML algorithms, including Multinomial Logistic Regression, Decision Tree, Random Forest, Support Vector Machine (SVM), K-Nearest Neighbors (KNN), XGBoost, Naïve Bayes, Ranger, and LogitBoost, are evaluated using 5-fold cross-validation. Performance metrics, including accuracy, sensitivity, specificity, precision, F1-score, false discovery rate (FDR), and area under the curve (AUC), are considered to compare the effectiveness of the models. Results indicate that among all ML models, Random Forest emerges as the top-performing model, achieving the highest accuracy (89.1%), macro-AUC (0.970), and weighted-AUC (0.968), and balanced performance across all classes. Blood Sugar is identified as the most critical predictor, followed by Systolic Blood Pressure and Age. Heart Rate and Body Temperature contribute minimally. The findings highlight the potential of ML techniques in enhancing the early detection of maternal health risks, thereby enabling timely interventions to improve healthcare outcomes.

International Journal of Statistical Sciences, Vol. 25(2), November, 2025, pp 59-80