Fetal Health Classification Using Machine Learning: Impact of Data Transformation and Balancing Methods

Health issues regarding mothers and fetuses have long been an important concern worldwide, especially in middle and low-income countries. Both the fetus and mother may experience severe health issues, which often lead to maternal mortality if the conditions for the fetus inside the mother’s womb are not suitable. Many medical experts use Cardiotocography (CTG) technology to anticipate the fetus’s heart rate to minimize those health problems. This study classifies fetal health from the CTG dataset using two ensemble classifiers, Random Forest (Bagging) and AdaBoost (Boosting) and two baseline classifiers, Gaussian Naive Bayes and Logistic Regression. This research further investigates the performance of these different models on three data transformation methods Discretization, Rank and Z-Score combined with PCA. Repeatedly, this study addresses the class imbalance issues using SMOTE and SMOTE-Tomek methods. In performance analysis, evaluating through 10-fold cross-validation it was found that Random Forest provided the best performance (97.58% accuracy) when both Z-Score (data transformation) and SMOTE-Tomek (data balancing technique) were used.