Classification of Maternal Health Risks Using Machine Learning Methods

Maternal health deaths and complications are a global concern. To reduce the mortality rate (~800 deaths/day), artificial intelligence/machine learning (AI/ML) techniques can be utilized by training the machine to develop predictive models. In this study, data was borrowed from UCI Machine Learning Repository and IBM Watson studio was used where predictive models were analyzed for their prediction accuracy, F1 score, recall, and feature importance, identifying the ones that played a significant role in model predictions. These included the XGB Classifier, Random Forest Classifier, Gradient Boosting Classifier, Extra Trees Classifier, and Snap Decision Tree Classifier. Depending on the algorithm, the model accuracies varied from 58.7% to 82.6%. Models were also developed using IBM Watson Jupyter Notebook in which the XGBoost, Random Forest, K-Nearest Neighbors, Support Vector Machine, and Logistic Regression, with accuracies ranging from 66% to 81.3%. It was found that the most important features were blood sugar and systolic blood pressure, while the age and heart rate were the least important. In addition, a website based on this research was developed using Python, Flask, HTML, and CSS where users, by providing input data, can run the model to determine the associated maternal health risk. The website analysis only provides an indication without any medical backup and demonstrates future AI/ML medical applications. It is anticipated that by making machine learning based predictive models more effective, thousands of patients can be saved through early diagnosis.