Ensemble machine learning framework for predicting maternal health risk during pregnancy

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Public Health

Background:

• Maternal health risks, including hypertension and glucose issues, contribute to pregnancy complications.
• Early identification and monitoring of these risks are crucial for reducing adverse outcomes.
• Existing methods may lack the precision needed for timely and effective intervention.

Purpose of the Study:

• To develop and implement a machine learning framework for predicting Maternal Health Risk (MHR) factors.
• To identify and classify maternal health risks using real-world datasets.
• To improve the accuracy of predicting high-risk pregnancies.

Main Methods:

• Developed the Quad-Ensemble Machine Learning framework (QEML-MHRC).
• Integrated multiple Machine Learning (ML) models with four ensemble ML techniques.
• Conducted nineteen training and testing experiments on data from maternity hospitals and clinics.

Main Results:

• Exploratory data analysis identified high blood pressure, low blood pressure, and high blood sugar as significant risk factors.
• The QEML-MHRC framework demonstrated outstanding predictive performance across all risk classes.
• The "HR" (High Risk) class achieved 90% prediction accuracy; Gradient Boosting Trees (GBT) with ensemble stacking showed 0.86 performance across all classes.

Conclusions:

• The proposed QEML-MHRC approach effectively assesses maternal health risks, enabling early detection of high-risk pregnancies.
• Class-wise performance measurement enhances the understanding of risk distinctions (high, low, medium).
• This predictive model can aid medical experts in timely interventions, potentially saving lives and reducing complications.