Effective Macrosomia Prediction Using Random Forest Algorithm.

Area of Science:

• Obstetrics and Gynecology
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Macrosomia, a condition of excessive birth weight, presents a significant global health challenge.
• Current prediction relies on ultrasonography, which has limitations in accuracy.
• Developing advanced predictive models is crucial for improving maternal and infant outcomes.

Purpose of the Study:

• To develop and validate novel predictive models for macrosomia using a random forest approach.
• To enhance the sensitivity and specificity of macrosomia prediction compared to existing methods.
• To identify key maternal factors influencing macrosomia risk.

Main Methods:

• Utilized a large dataset from the Shandong Multi-Center Healthcare Big Data Platform (June 2017-May 2018).
• Constructed and compared a random forest model against a logistic regression model and traditional ultrasound.
• Analyzed influencing factors based on the average decrease in Gini coefficient.

Main Results:

• The random forest model achieved 91.7% sensitivity, 91.7% specificity, and a 95.3% area under the curve (AUC).
• Logistic regression yielded 56.2% sensitivity, 82.6% specificity, and a 72.0% AUC.
• Ultrasound showed 29.6% sensitivity and 97.5% specificity.

Conclusions:

• A maternal information-based random forest model offers accurate prediction of macrosomia during pregnancy.
• This model provides a foundation for developing efficient screening and diagnostic tools.
• The findings highlight the potential of machine learning in improving obstetric care.