Maternal Health Risk Prediction with Machine Learning Methods

Abstract

Ensuring maternal well-being throughout pregnancy is a critical concern that directly impacts both the mother and the unborn child. Accurate assessment of pregnancy risks is vital for timely medical intervention and appropriate allocation of care. The goal of this undertaking is to develop a web application that incorporates a sophisticated machine-learning model. The selection of this model was based on a comprehensive comparison of different approaches. The primary function of this model is to classify pregnancy risk levels into three categories: high, medium, or low. This classification will be based on crucial vital signs obtained from the expectant mother. The dataset used contains 1014 samples gathered using an Internet of Things based risk monitoring system. This determination relies on essential input parameters including the data collected includes the individual's age, readings of both systolic and diastolic blood pressure, blood glucose levels, body temperature, and pulse rate. Leveraging a dataset comprising historical patient records encompassing the aforementioned vital signs and corresponding risk assessments provided by medical experts, the model will undergo training to discern intricate patterns and interrelationships between the input variables and the corresponding risk categories. After comparing accuracy by the classifiers Logistic Regression(58%), Random Forest(80%), Support Vector Machine(58%), Decision Tree(58%), Naive Bayes(60%), K-Nearest Neighbors(64%), Gradient Boosting(72%), AdaBoost(74%), and ensemble of Random Forest, AdaBoost and Gradient Boosting with hard voting(78%) and soft voting(81.5%), best accuracy is achieved with an ensemble of Random Forest, Gradient Boosting and AdaBoost with Decision trees with soft voting. The accuracy of this classifier is 81.5%. It empowers the medical professionals to allocate resources judiciously and administer targeted care to pregnant patients. By precisely forecasting pregnancy risk levels, this innovative machine-learning model stands to considerably enhance maternal and fetal healthcare outcomes. It enables prompt interventions and personalized care strategies that have the potential to significantly elevate the quality of healthcare provided.