A Decision Tree-Based Classification of Fetal Health Using Cardiotocograms

Child survival, in the Sustainable Development Goals (SDG), endeavors to increase the child survival rate. Neonatal mortality rate is aimed to be dropped to 1.2% and under-5 mortalities to 2.5% by 2030. Cardiotocographs (CTGs) are used for fetal health assessment during pregnancies with risk of complications by continuously recording the heart rate, movements, uterine contractions, etc. of the fetus. This allows medical practitioners to take measures to improve the outcomes for the mothers and their babies. However, there are no direct methods to determine the fetal state in CTGs. The interpretation of the fetal health state is based on the medical practitioner's skill and experience. This study uses the CART (Classification and Regression Tree) decision tree learning algorithm with Gini index to analyze the dataset and create a model that can predict which class (normal, suspect, or pathological) the new data belongs to. The dataset consists of 2126 CTG recordings that were extracted from various exams. The samples were split using a 6:4 train-test ratio. The best model was evaluated using feature selection and scaling methods. The accuracy yielded by the best model was 93.65%. Due to its high accuracy, a Decision Tree model could be implemented as a medical decision support system in CTGs fetal state classification.