Predicting Labor and Delivery Complications: Dual Application of Logistic Regression and Supervised Machine Learning Algorithms

Abstract

Abstract Background Prediction involves the use of data to learn, build knowledge, and improve predictive capacity through time from historical data to forecast future events. Predicting labor and delivery complications at an early stage could save the mother and baby from health challenges even death. This study aimed to identify determinants of and predict labor and delivery complications using machine learning techniques. Methods Data were collected using a data extraction sheet adopted from the Federal Ministry of Health Integrated antenatal, labor, delivery, and postnatal care card from Ayder Comprehensive Specialized Hospital, Ethiopia from April to July 2020. These samples were grouped using an 80% by 20% ratio on stratified outcome variables into training and test datasets. Descriptive, bivariate, and multivariate regression analyses were performed using Statistical Package for Social Science (SPSS). The synthetic minority oversampling technique (SMOTE) was used to balance the training dataset. Python and scikit learn were utilized to implement extreme gradient boosting (XGB), random forest, decision tree (DT), support vector machine (SVM), and K-nearest-neighbors (KNN) to develop predictive models for predicting labor and delivery complications. The confusion matrix, accuracy, precision, recall, receiver operating characteristics (ROC) curve, and F1-score test were used to compare the classification algorithms' prediction performance. Results A total of 320 (16%) mothers experienced labor and delivery complications. Models developed with KNN, SVM, random forest, DT, and XGB predicted the occurrence of labor and delivery complications with accuracy levels of 82%, 82%, 80%, 82% and 85%, respectively. The model developed using XGB scored the highest accuracy level. Conclusions Age, history of hypertension, history of preeclampsia, history of abortion, vaginal bleeding in the current pregnancy, history of diabetes mellitus, presentation of fetus, and Rh status were found to be determinants of labor and delivery complications. The model developed using the XGB algorithm has performed better in terms of predictive performance.