The Impact of Data Balancing on the Classifier's Performance in Predicting Cesarean Childbirth

Abstract

The number of cesarean sections delivered world-wide is increasing at an alarming rate. It has a negative influence on the health of both mother and child, as well as on the economy. This problem can be effectively alleviated with the early-stage prediction of cesarean section. We implemented Logistic Regression (LR) and Multilayer Perceptron (MLP) to predict cesarean section from secondary data in this study. The necessary features are chosen using the Chi-square test, univariate feature selection, and the forward feature selection approach. Several imbalance data handling techniques are used to mitigate the effects of imbalanced data on classifiers. The accuracy of imbalance data prediction is satisfactory for both methods, but the classification metrics for the two classes are varied. Both models show a significant outcome by using specified features and SMOTE oversampling techniques. The pipeline shows 93% accuracy with greater than 93% precision, recall, and F1 score for both classes of LR prediction, and 95% accuracy with greater than 95% precision, recall, and F1 score for both classes of MLP prediction.