Predicting Maternal Health Risks and Fetal Impacts using Machine Learning

Abstract

The research study applies machine learning algorithms to produce simultaneous forecasts about maternal and fetal health in an innovative integrated approach. We implemented and deployed two primary models including one for fetal health classification and another for pregnant health prediction. The research objective involves identifying maternal health risks followed by assessing their influence on fetal health results. Our system operates through a user-friendly web-based platform based on Streamlit development that enables medical practitioners to easily interact with and review analytical results. The fetal health categorization model analyzes full cardiotocography (CTG) records yet the pregnancy risk prediction model works with crucial maternal health measurements. The integrated approach demonstrates its ability to deliver rapid and exact risk assessments through the findings that could boost maternal care choices while improving health results for mothers alongside their newborns.

Introduction

Global Context & Motivation:
Maternal and neonatal health remains a critical issue, especially in low-resource settings, where over 295,000 women died from pregnancy-related causes in 2017, and 2.5 million newborns died within 30 days in 2018. These alarming figures highlight the need for intelligent, early-risk detection systems to improve prenatal care and reduce preventable deaths.

Proposed System:
The study introduces a dual-model machine learning (ML) system integrated into a web platform (via Streamlit) to predict:

1. Maternal health risks using blood pressure, glucose, and other vitals.

2. Fetal health conditions based on Cardiotocography (CTG) data.

Both models use Gradient Boosting, selected after comparison with Logistic Regression, K-Nearest Neighbors (KNN), and Random Forest algorithms. The fetal model incorporates SMOTE to address class imbalance.

Methodology Highlights:

• Datasets:

  • Maternal Health Dataset (1,014 entries)

  • Fetal Health Dataset (2,126 CTG records)

• Preprocessing:

  • Outlier removal (e.g., unscientific heart rate values)

  • Imputation (mean for continuous, mode for categorical)

  • Feature scaling (Min-Max for maternal, Z-score for fetal data)

  • SMOTE for fetal dataset to balance classes

• Model Evaluation:

  • Train-Test split (80/20)

  • 5-fold cross-validation

  • Evaluation metrics: Accuracy, Precision, Recall, F1-score, AUC

• Deployment:

  • Models deployed using joblib

  • Real-time predictions via Streamlit web interface

  • Visual tools include heatmaps and feature importance metrics

  • SHAP explainability planned for transparency

Key Results:

• Maternal Health Model (Gradient Boosting):

  • Accuracy: 84.2%

  • F1-score: 84.1%

  • Outperformed all other algorithms; Logistic Regression performed the worst (63.5% accuracy)

• Fetal Health Model (Gradient Boosting):

  • Accuracy, Precision, Recall, F1-score: 93%

  • AUC: 0.97

  • High specificity: minimized false alarms for normal cases

• System Strengths:

  • Rapid inference (<0.5 sec)

  • High interpretability and clinical usability

  • Effective even with class imbalance and nonlinear data

• Limitations:

  • Trained on retrospective data; real-world validation is needed

  • Risk of bias due to unbalanced patient representation

  • Ethical considerations such as fairness and trust must be addressed further

Literature Review Insights:

• Various studies (Kalita, Mohanty, Shifa, Bajaj) have used ML for maternal/fetal health.

• Techniques like blockchain integration, ensemble learning, and SMOTE have shown improved outcomes.

• Gradient Boosting and Random Forest frequently emerged as top performers in previous work.

Conclusion

The study shows that an explanatory procedure combining machine learning solutions effectively detects maternal health perils alongside fetal conditions in real-time. Tests conducted on four supervised learning algorithms led to Gradient Boosting selection as its performance exceeded all others during metric evaluations. Our system built a reliable clinical decision support through its methodical approach to modeling combined with strong data pre-processing protocols and statistical model evaluation. The dual-model system reached success by deploying a Streamlit-based web application which generates instant interpretable predictions to enable healthcare staff to address potential medical issues promptly. The unified system successfully meets the research requirement for delivering an ML solution which enhances prenatal care quality while detecting conditions early. The experimental success of this technique encounters limitations because it depends heavily on past data records while facing possible age-related biases. The upcoming phase of research development will aim to connect real-time hospital capabilities with diverse live dataset applications while also implementing explainable ML tools to maintain fair and transparent prediction methods. The research shows how machine learning methods can connect important gaps in maternal and fetal healthcare to deliver proactive customized diagnoses in order to improve health results.

References

[1] K. P. Kalita, S. K. Chettri and R. K. Deka, \"A Blockchain-based Model for Maternal Health Information Exchange and Prediction of Health Risks using Machine Learning,\" 2023 International Conference on Intelligent and Innovative Technologies in Computing, Electrical and Electronics (IITCEE), Bengaluru, India, 2023, pp. 1184-1189, doi: 10.1109/IITCEE57236.2023.10090997 [2] R. Mohanty et al., \"Comparitive Analysis of Machine Learning Models for Prediction of Fetal Health,\" 2024 International Conference on Advancements in Smart, Secure and Intelligent Computing (ASSIC), Bhubaneswar, India, 2024, pp. 1-5, doi: 10.1109/ASSIC60049.2024.10508030 [3] H. A. Shifa, M. U. Mojumdar, M. M. Rahman, N. R. Chakraborty and V. Gupta, \"Machine Learning Models for Maternal Health Risk Prediction based on Clinical Data,\" 2024 11th International Conference on Computing for Sustainable Global Development (INDIACom), New Delhi, India, 2024, pp. 1312-1318, doi: 10.23919/INDIACom61295.2024.10498822. [4] D. Bajaj, R. Kumari and P. Bansal, \"Risk Level Prediction for Maternal Health Using Machine Learning Algorithms,\" 2023 International Conference on Communication, Security and Artificial Intelligence (ICCSAI), Greater Noida, India, 2023, pp. 405-409, doi: 10.1109/ICCSAI59793.2023.10421156. [5] M. Assaduzzaman, A. A. Mamun and M. Z. Hasan, \"Early Prediction of Maternal Health Risk Factors Using Machine Learning Techniques,\" 2023 International Conference for Advancement in Technology (ICONAT), Goa, India, 2023, pp. 1-6, doi: 10.1109/ICONAT57137.2023.10080700. [6] A. I. Reddy and W. D. Priya, \"An Efficient Fetal Health Prediction System Using Enhanced XGB Classifier Compared Over K-Nearest Neighbor Classifier with Less Complexity and Improved Accuracy,\" 2023 Intelligent Computing and Control for Engineering and Business Systems (ICCEBS), Chennai, India, 2023, pp. 1-5, doi: 10.1109/ICCEBS58601.2023.10448547. [7] A. P, T. D, V. M J, V. B C and V. G, \"Fetal Health Prediction,\" 2024 Second International Conference on Advances in Information Technology (ICAIT), Chikkamagaluru, Karnataka, India, 2024, pp. 1-5, doi: 10.1109/ICAIT61638.2024.10690588. [8] M. Ramla, S. Sangeetha and S. Nickolas, \"Fetal Health State Monitoring Using Decision Tree Classifier from Cardiotocography Measurements,\" 2018 Second International Conference on Intelligent Computing and Control Systems (ICICCS), Madurai, India, 2018, pp. 1799-1803, doi: 10.1109/ICCONS.2018.8663047. [9] M. Moutaib, M. Fattah, Y. Farhaoui and B. Aghoutane, \"Machine Learning in Fetal Health: Improving ECG Analysis with Random Forest,\" 2024 International Conference on Circuit, Systems and Communication (ICCSC), Fes, Morocco, 2024, pp. 1-5, doi: 10.1109/ICCSC62074.2024.10617320.

Copyright

Copyright © 2025 Deepali Deshpande, Shlok Bam, Shreyas Chulliyan, Samruddhi Shinde, Lavanya Veni. 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.