A MACHINE LEARNING-BASED FRAMEWORK FOR PRETERM BIRTH PREDICTION USING MATERNAL DATA IN RURAL HEALTHCARE SETTINGS

Abstract

Preterm birth (PTB) is a major contributor to neonatal mortality and long-term health complications, particularly in rural and resource-limited regions where access to advanced diagnostic facilities is restricted. Early prediction of PTB is critical for timely intervention; however, conventional clinical approaches are often costly and less accessible. This study presents a machine learning-based predictive framework for assessing preterm birth risk using maternal and obstetrical data collected from a rural healthcare center. The proposed approach incorporates data preprocessing, discretization, and an entropy-based feature selection technique to identify the most relevant maternal risk factors. Three classification algorithms—Decision Tree (DT), Logistic Regression (LR), and Support Vector Machine (SVM)—are evaluated using performance metrics such as accuracy, sensitivity, and specificity. Due to class imbalance in the dataset, the Synthetic Minority Oversampling Technique (SMOTE) is applied to improve model learning. Experimental results demonstrate that SVM achieves the highest accuracy of 86.1% on the original dataset, which significantly improves to 90.9% after dataset balancing. Additionally, sensitivity increases to 89.1%, indicating enhanced capability in identifying high-risk pregnancies. Comparative analysis confirms that SVM outperforms DT and LR in predictive performance. The results highlight the effectiveness of combining feature selection and data balancing with machine learning models for accurate PTB prediction. The proposed framework offers a cost-effective and reliable solution for early risk detection, making it highly suitable for deployment in rural healthcare environments to support clinical decision-making and improve maternal and neonatal outcomes.