Towards the Development of a Predictive Machine Learning Model for Maternal Mortality

Abstract

Data from World Health Organization (WHO) on maternal mortality show that Sub-Saharan African countries account for over 50% of the records, with Nigeria having one of the highest maternal mortality rates in the world. This study explores the intervention of Artificial Intelligence in health care by proposing a predictive ensemble model to classify maternal health risks.  Relevant Maternal Health Risk dataset was sourced from the University of California (UCI) machine learning repository. The dataset was preprocessed and further split into training and validation data. Various supervised machine learning (ML) algorithms were thereafter trained and validated via the training and validation datasets respectively. Data analysis, visualization and entire model development was carried out using the Google Colaboratory cloud framework. Seven base classifiers were applied to the dataset namely, Linear Support Vector Machine, Gaussian Naïve Bayes, Multilayer Perceptron, Decision Tree, Random Forest, Gradient Boosting Decision Tree and Extreme Gradient Boosting. An ensemble model was thereafter developed from the seven classifiers to further enhance the predictive performance. Results showed that Random Forest had 86% precision and Extreme Gradient Boosted Decision Tree had 85% as the best performing algorithms. The least performing algorithm is the Gaussian Naïve Bayes with 61% precision. Results also showed the ensemble model with a higher precision score of 92% which is higher than that of Random Forest (the best forming base model), before stacking was executed. The accuracy score of the ensemble model is also higher than it was recorded (84% for Random Forest) before stacking.