Related Experiment Videos
Machine learning-based fetal health prediction and development of smart web application
Chetan Puri1, K T V Reddy2, Pradnyawant M Gote3
1Department of Computer Science and Engineering, Faculty of Engineering and Technology, Datta Meghe Institute of Higher Education and Research (DU), Wardha, Maharashtra, India.
Frontiers in Artificial Intelligence
|July 23, 2026
Summary
A machine learning model accurately predicts fetal health using cardiotocography (CTG) data, improving upon subjective manual interpretations. The LightGBM model achieved 96.03% accuracy, offering a potential clinical decision-support tool for pregnancy risk identification.
Area of Science:
- * Obstetrics and Gynecology
- * Medical Informatics
- * Machine Learning in Healthcare
Background:
- * Manual interpretation of cardiotocography (CTG) signals for fetal health monitoring is subjective and inconsistent.
- * Early identification of pregnancy-related risks necessitates objective and reliable fetal health assessment.
- * Existing methods lack the precision required for consistent clinical decision-making.
Purpose of the Study:
- * To develop and evaluate a machine learning framework for objective fetal health classification.
- * To compare the performance of eleven different machine learning and neural network models.
- * To identify key clinical features influencing fetal health classification.
Main Methods:
- * Preprocessing included duplicate removal, normalization, SMOTEENN for class balancing, VIF for multicollinearity, and Kruskal-Wallis for feature selection.
- * Eleven models (Logistic Regression, KNN, SVM, Decision Tree, Random Forest, Gradient Boosting, AdaBoost, XGBoost, LightGBM, MLP, DNN) were trained and compared.
- * SHAP analysis was employed for model explainability.
Main Results:
- * The LightGBM model demonstrated superior performance with 96.03% accuracy and 99.02% ROC-AUC.
- * Key features identified by SHAP include abnormal short-term variability and fetal heart rate accelerations.
- * The model achieved high scores across various metrics including balanced accuracy, macro F1-score, Cohen's Kappa, and MCC.
Conclusions:
- * A machine learning framework, particularly LightGBM, offers a highly accurate method for fetal health classification.
- * The developed model can serve as a valuable clinical decision-support tool for real-time fetal health prediction.
- * Explainable AI insights highlight critical features for understanding fetal distress.
