Evaluating the Effectiveness of LightGBM for Fetal Health Prediction

The health of the fetus is essential for assuring that babies are born healthy. A mother’s fetus must be in excellent health. The categorization of fetal health is essential for assessing the well-being and development of unborn infants. Despite this, numerous findings continue to indicate a high neonatal mortality rate. During pregnancy and labor, the Cardiotocography (CTG) is a common method for monitoring fetal health. It measures fetal heart rate and uterine contractions, allowing healthcare professionals to evaluate the baby’s health and identify potential problems early on. However, interpretation of CTG data remains subjective, resulting in potential misdiagnoses of fetal health. This study seeks to reduce infant mortality rates caused by delivery complications by employing machine learning technology to assess a baby’s health based on Cardiotocography test results. In this study, we evaluated the performance of three machine learning algorithms - XGBoost, Random Forest, and LightGBM - in predicting fetal health based on Cardiotocography data. We implemented various scenarios involving preprocessing techniques, feature selection methods, and hyperparameter tuning to enhance the model’s prediction accuracy. The scenarios are without preprocessing, feature selection using Information Gain, feature selection using K-Best, Feature selection using both feature selection algorithm, without hyperparamter tuning and hyperparamter tuning with Grid Search. The result showed the use of all techniques combined proved beneficial, resulting in accuracies of 96% for XGBoost, 95% for Random Forest, and 97% for LightGBM.