Comparison of XGboost, Extra Trees, and LightGBM with SMOTE for Fetal Health Classification

Abstract

Cardiotocography (CTG) is widely used by obstetricians to physically access the condition of the fetus during pregnancy. This can provide data to the obstetrician about fetal heart measurements and uterine duration which helps determine whether the fetus is pathological or not. Determining the pathological classification or not can be done using machine learning methods. In this research, there is a problem of unbalanced data or data imbalance. To overcome data instability, testing using SMOTE is used. Then a comparison was made with the classifications, namely XGboost, Extra Trees and LightGBM. XGboost, Extra Trees and LightGBM testing results using SMOTE obtained the best results at 91.52% accuracy, 90.49% recall and 89.12% f1-score produced by LightGBM. Meanwhile, the best results were 89.07% precision and AUC 0.9800 produced by Extra Trees.