Machine learning-based approach for predicting low birth weight.

Area of Science:

• Perinatal health
• Medical informatics
• Predictive analytics in obstetrics

Background:

• Low birth weight (LBW) is a significant risk factor for infant mortality and adverse newborn health outcomes.
• Accurate prediction of LBW is crucial for timely intervention and improved neonatal care.
• The study addresses the need for effective predictive tools in maternal and neonatal health.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting low birth weight (LBW).
• To compare the diagnostic performance of various statistical learning models in identifying LBW.
• To identify key predictors of LBW in singleton pregnancies.

Main Methods:

• Utilized data from the Iranian Maternal and Neonatal Network (IMaN Net) database (January 2020 - January 2022).
• Included singleton pregnancies >24 weeks gestation; excluded multiple pregnancies and fetal anomalies.
• Evaluated eight machine learning models including deep learning, random forest, and extreme gradient boost, using AUROC, accuracy, precision, recall, and F1 score for performance assessment.

Main Results:

• A total of 8853 deliveries were analyzed, with a low birth weight (LBW) incidence of 14.5% (1280 cases).
• Deep learning (AUROC: 0.86), random forest (AUROC: 0.79), and extreme gradient boost (AUROC: 0.79) demonstrated superior performance.
• The extreme gradient boost model achieved the highest diagnostic performance with an accuracy of 0.79, precision of 0.87, recall of 0.69, and F1 score of 0.77.

Conclusions:

• The extreme gradient boost model demonstrated strong predictive capability for low birth weight (LBW).
• Gestational age and a history of prior LBW were identified as critical predictors.
• Further research is warranted to definitively establish the optimal machine learning model for LBW prediction.