Application of machine learning algorithms to predict factors of cesarean delivery among women in Rwanda

Abstract

Cesarean delivery is a surgical technique typically performed when vaginal childbirth would pose risk to the health of both the mother and the baby . From 1990 to 2021, the rate of cesarean deliveries increased from 7% to 21.1% of all global deliveries. Cesarean deliveries are becoming increasingly common in Sub-Saharan African , just asthey are in Rwanda in 2021, they accounted for 15.6% of all births. Machine learning (ML) techniques have not yet been applied in Rwanda to predict cesarean delivery and identify the factors contributing to its increasing prevalence. As result, machine learning algorithms were employed to predict cesarean deliveries among women, which served as the project’s primary objective. This initiative aims to enhance maternal and newborn safety by reducing unnecessary cesarean deliveries during labor and after.This study used a cross-sectional secondary data from the Rwanda demographic health survey (RDHS 2019-2020) to predict cesarean deliveries. To achieve the main research goal, five ML algorithms, a namely logistic regressions, decision trees, random forests, K nearest neighbor (KNN) classifier and support vector machine (SVM) were applied. Evaluation metrics such confusion matrix, precision, accuracy, area under curve score, and F1 score were used to determine the best cesarean delivery prediction model. Multivariable logistic regression results show that, several factors linked to higher cesarean delivery likelihood in Rwandan women. These factors include birth order, number of antenatal care visits, birth weight, mother’s education level, short rapid breaths of mother during labor, mother’s wealth index, place of residence, mother’s current age, age of mother at first birth, number of terminated pregnancies , and the twin status of the child. In terms of predictive models for cesarean delivery created using machine learning techniques, the random forest model emerged as the most effective model achieving an accuracy rate of 83.0% and an area under the curve (AUC) score of 90.2%. The decision tree model followed closely with an accuracy rate of 79% and AUC score of 80.7%. The k-nearest neighbor model achieved an accuracy rate of 75% and AUC score of 81.4%. Meanwhile, the support vector machine model exhibited an accuracy rate of 63% and AUC score of 63%. Lastly, the logistic regression model had the lowest performance , with an accuracy of 61% and AUC score of 66.2%, when compared to the other predictive models used in the study. Our research highlights various factors driving the increase in cesarean sections among Rwandan women. Therefore, healthcare policies should explore alternative strategies to ensure women can access essential cesarean deliveries as advised by a doctor.