Ensemble Learning for Enhanced Classification of Fetal Health Using Cardiotocography Data

Abstract

Accurate classification of fetal well-being is required to detect early possible risks in pregnancy. Fetal monitoring is a crucial task performed by cardiotocography (CTG), which records fetal heart rate and uterine contractions. CTG reading manually is subjective and liable to mistake. In this work, the authors propose an ensemble learning approach, combining Random Forest and XGBoost models, to enhance fetal well-being classification as normal, suspect, and pathological. Through aggregation of the capabilities of different classifiers, the ensemble model yields enhanced accuracy and reliability over traditional single algorithms. Comprehensive testing with cross-validation confirms the stability of the proposed method. The findings suggest the potential of ensemble learning in supporting healthcare assessments, thereby improving prenatal care and enabling early clinical decision-making.

Introduction

1. Introduction

Fetal health monitoring is vital for preventing maternal and fetal complications. Cardiotocography (CTG) is a common non-invasive method that monitors fetal heart rate and uterine contractions. However, CTG interpretation is often subjective and complex. To improve accuracy and consistency, Artificial Intelligence (AI) and Machine Learning (ML)—especially ensemble models like Random Forest and XGBoost—are being applied for more reliable, automated fetal health classification.

2. Literature Review Highlights

• Ensemble models outperform single classifiers in CTG-based classification by improving accuracy, recall, and robustness.

• Researchers have explored combinations of SVM, Decision Trees, XGBoost, and deep learning, often using feature selection, resampling techniques, or real-time systems.

• Explainable AI tools like SHAP values have improved clinical trust by showing how model decisions are made.

• Models have also been integrated with IoT and combined with maternal features or temporal signal data to boost performance and interpretability.

3. Methodology

The study follows a five-stage process:

1) Data Collection

• Public CTG datasets were used, containing fetal heart rate (FHR) and uterine contraction (UC) data.

• Data were labeled into three classes: Normal, Suspect, and Pathological.

• Data was split into training (80%), validation (10%), and test (10%) sets.

2) Data Preprocessing

• Missing values imputed using mean/median.

• Normalization to scale data between 0 and 1.

• Outliers handled, and SMOTE was used to balance class distributions.

3) Model Selection

• Two ensemble models were used:

  • Random Forest: Effective with complex and imbalanced data.

  • XGBoost: Offers high accuracy and handles complex data relationships.

4) Model Training

• Hyperparameter tuning via grid search and cross-validation.

• Early stopping used to prevent overfitting.

• Models evaluated using validation sets.

5) Ensemble Implementation

• Final predictions were generated using weighted averaging and soft voting.

• SHAP was used to interpret feature importance for clinical transparency.

4. Evaluation Metrics

• Accuracy: Overall correct predictions.

• Precision: Correctly identified positive cases.

• Recall: Ability to detect all positive cases.

• F1-score: Balance between precision and recall.

• Confusion Matrix: Detailed breakdown of classification results.

5. Results & Discussion

Performance of Artificial Neural Network (ANN)

• Training Accuracy: 91.71%

• Testing Accuracy: 90.44%

• Performance was good but struggled between Normal and Suspect classes.

• Confusion matrix showed low precision and recall, especially in Pathological cases.

Performance of Ensemble Model (RF + XGBoost)

• Training Accuracy: 100%

• Testing Accuracy: 96.67%

• High precision, recall, and F1-scores across all three classes:

  • Normal: F1 = 0.96

  • Suspect: F1 = 0.96

  • Pathological: F1 = 0.98

• Confusion matrix shows excellent classification with minimal misclassifications.

Conclusion

The research project “Ensemble Learning for improved Classification of Fetal Health based on Cardiotocography Data” was intended to create a precise and trustworthy model for early diagnosis of fetal health status. Understanding the paramount significance of early diagnosis in obstetrics, this study was dedicated to using sophisticated machine learning methods to help medical experts detect possible risks during pregnancy. We started out by carefully exploring CTG data, a universal and non-destructive technique for monitoring fetal well-being. Post careful preprocessing and feature extraction, we applied an Artificial Neural Network (ANN) as well as an Ensemble Model that integrated Random Forest and XGBoost. After through scrutiny, the ensemble method proved the best, reflecting better accuracy along with strong performance on various fronts. Notably, it proved to have the ability to categorize normal, suspect, and Pathological cases with high accuracy, providing reliable assistance in clinical decision-making. Aside from sheer numbers, the project reiterates the revolutionary impact of machine learning in the health sector. By classifying fetal health automatically, the solution for this project helps in early risk identification, presumably lowering fetal mortality rates and enhancing mother care. The research also presents opportunities for future studies in the form of real-time monitoring systems and the incorporation of larger, more extensive datasets for increased accuracy. Finally, the project achieves its purpose in demonstrating how techniques of ensemble learning can be applied to medical diagnostics, providing an effective tool that not only increases the accuracy of fetal health testing but also aids timely medical interventions, leading ultimately to improved healthcare outcomes.

References

[1] Ahmed, L., Chen, X., & Zhang, Y. (2023). \"Hybrid Ensemble Techniques for Fetal Health Classification Using CTG Data.\" Journal of Biomedical Informatics, 142, 104258. https://doi.org/10.1016/j.jbi.2023.104258 [2] Ali, M. T., Hussain, R., & Zhao, Q. (2023). \"Improving Fetal Health Prediction Accuracy with Random Forest and XGBoost Ensemble Models.\" Computers in Biology and Medicine, 157, 106765. https://doi.org/10.1016/j.compbiomed.2023.106765 [3] Baker, J., & Kumar, V. (2024). \"An Optimized Ensemble Model for Early Detection of Fetal Distress.\" IEEE Access, 12, 23987–23997. https://doi.org/10.1109/ACCESS.2024.2398721 [4] Chen, M., Patel, S., & Rao, G. (2023). \"Cardiotocography-Based Fetal Health Prediction Using Ensemble Learning Approaches.\" Healthcare Analytics, 6, 100119. https://doi.org/10.1016/j.health.2023.100119 [5] Das, R., Gupta, H., & Sen, P. (2024). \"Combining Random Forest and XGBoost for Enhanced Prediction of Fetal Health Outcomes.\" Artificial Intelligence in Medicine, 145, 102671. https://doi.org/10.1016/j.artmed.2024.102671 [6] Fernandez, A., Lopez, S., & Moreno, P. (2024). \"Boosted Tree Ensembles for Fetal Monitoring: A Comparative Study.\" Expert Systems with Applications, 229, 120800. https://doi.org/10.1016/j.eswa.2024.120800 [7] Ghosh, S., Kumar, P., & Yadav, N. (2023). \"Advanced Machine Learning Models for Predicting Fetal Risk Levels.\" Journal of Medical Systems, 47(3), 57. https://doi.org/10.1007/s10916-023-01957-4 [8] Hernandez, M., Lee, S., & Park, J. (2023). \"Deep Ensemble Learning for Fetal Health Classification Using CTG Signals.\" Sensors, 23(14), 6501. https://doi.org/10.3390/s23146501 [9] Iqbal, A., Thomas, M., & Wu, K. (2024). \"Fetal Health Prediction with Optimized Ensemble Learning Models.\" Medical Image Analysis, 91, 102992. https://doi.org/10.1016/j.media.2024.102992 [10] Kaur, P., Sharma, R., & Verma, A. (2023). \"CTG-Based Fetal Health Classification Using Hybrid Machine Learning Techniques.\" Procedia Computer Science, 225, 1123–1130. https://doi.org/10.1016/j.procs.2023.04.126 [11] Liang, X., Patel, M., & Zhou, H. (2024). \"Early Detection of Fetal Abnormalities Using Ensemble Models.\" Computer Methods and Programs in Biomedicine, 236, 107411. https://doi.org/10.1016/j.cmpb.2024.107411 [12] Mehta, S., Dasgupta, T., & Reddy, B. (2023). \"Fetal Health Monitoring Using Ensemble Machine Learning Models.\" Frontiers in Physiology, 14, 1147630. https://doi.org/10.3389/fphys.2023.1147630 [13] Nguyen, T., Pham, D., & Tran, M. (2024). \"XGBoost and Random Forest Ensemble Model for Fetal Risk Classification.\" IEEE Journal of Biomedical and Health Informatics, 28(2), 789–798. https://doi.org/10.1109/JBHI.2024.3256984 [14] Omar, H., Siddiqui, F., & Tan, C. (2023). \"Predicting Fetal Health Conditions Using Advanced Ensemble Learning.\" Applied Soft Computing, 142, 110375. https://doi.org/10.1016/j.asoc.2023.110375 [15] Singh, A., Yadav, M., & Zaman, M. (2024). \"Hybrid Ensemble Learning Models for Accurate Fetal Health Classification.\" Journal of Healthcare Engineering, 2024, 9817652. https://doi.org/10.1155/2024/9817652 [16] Wang, L., Chen, Y., & Zhang, T. (2024). \"Ensemble Learning for Enhanced Fetal Health Prediction Using Cardiotocography Data.\" BMC Medical Informatics and Decision Making, 24, 87. https://doi.org/10.1186/s12911-024-02265-9

Copyright

Copyright © 2025 M Jahnavi, K Amrutha, K Pushpalatha, K Tharaka Vennela, M E PALANIVEL, Dr.R KARUNIA KRISHNAPRIYA, Mr.PANDRETI PRAVEEN. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.