Prediction of Maternal Health Risk based on Physiological Indicators

The paper aims to find out which factors mostly lead to the health risks of pregnant women and search for the appropriate model for prediction. The sample number is 1012. Boxplots and correlation heatmap are used for data visualization. There are six variables and the risk level given in the data set is considered as a factor to describe the condition of pregnant women. Level is divided into 3 levels: high, medium and low. Variable "BS", which refers to blood glucose levels in terms of a molar concentration, is the most important factor and systolic BP have the strongest correlation with diastolic BP after data visualization. Two models, random forest and logistic regression, are selected to calculate the estimate, std. error and z values of each variable. The results show that training set has 809 samples and testing set has 203 samples. Logistic regression uses 3 different models and model 2 and 3 proves to be the effective ones since variables in the two models are more significant than model 1. After data visualization process, variable "BS"proves to be the most important factor to the health risks of pregnant women. Random forest has a higher accuracy than logistic regression, so it turns out to be a more effective model to predict the high risk pregnant women. © 2023 Copyright held by the owner/author(s). Publication rights licensed to ACM.