A Machine Learning-Based Prediction Model for Preterm Birth in Rural India.

Area of Science:

• Obstetrics and Gynecology
• Machine Learning in Healthcare
• Data Science

Background:

• Preterm birth (PTB) is a significant global health concern, particularly in rural India.
• Existing clinical prediction models for PTB often lack optimal feature selection efficiency.
• There is a need for accurate and timely prediction of PTB to improve maternal and neonatal outcomes.

Purpose of the Study:

• To design and evaluate a machine learning model, the Risk Prediction Conceptual Model (RPCM), for predicting preterm birth.
• To develop a novel feature selection approach based on entropy for identifying key maternal risk factors.
• To enhance the accuracy of PTB prediction models by efficiently selecting relevant features from obstetrical datasets.

Main Methods:

• A literature review on PTB cases was conducted.
• Obstetrical data was collected from a rural Community Health Centre in Jharkhand, India.
• A feature selection method based on entropy was applied to identify significant maternal features.
• Three machine learning classifiers (Decision Tree, Logistic Regression, Support Vector Machine) were implemented for PTB prediction.

Main Results:

• The entropy-based feature selection identified crucial maternal indicators for PTB.
• The Support Vector Machine (SVM) classifier achieved the highest prediction accuracy of 90.9%.
• The performance of Decision Tree and Logistic Regression classifiers was also evaluated in terms of accuracy, specificity, and sensitivity.

Conclusions:

• The developed RPCM, utilizing an entropy-based feature selection and SVM classifier, demonstrates high accuracy in predicting preterm birth.
• This approach offers a promising tool for early identification of PTB risk in rural healthcare settings.
• The study highlights the potential of machine learning in addressing critical obstetric challenges in underserved regions.