Pregnancy Health Risk Prediction using Machine Learning and Deep Learning

---

Overview

Pregnancy health risk analysis is critical for ensuring the well-being of both mother and fetus. Timely assessment and intervention are essential to prevent complications. This repository presents an automated system for predicting pregnancy-related health risks using Machine Learning (ML), Deep Learning (DL), hybrid models, and ensemble techniques.

The study leverages both maternal and fetal health datasets and proposes a hybrid BiGRU-BiLSTM model that demonstrates superior performance. Additionally, explainable AI techniques are used to interpret model predictions and identify key contributing features.

DOI

https://doi.org/10.1109/ICISET62123.2024.10939599

Research Objectives

Objective ID	Description
O1	Predict maternal and fetal health risks
O2	Compare ML, DL, hybrid, and ensemble approaches
O3	Handle data imbalance and improve data quality
O4	Develop an interpretable and generalizable model

---

Datasets

Dataset	Description
Maternal Health	Clinical features related to maternal condition
Fetal Health	Indicators related to fetal well-being
Merged Dataset	Combined features for adaptability evaluation

---

Data Processing Techniques

Technique	Purpose
Data Preprocessing	Cleaning and preparation of raw data
SMOTE	Handling class imbalance
cGAN	Synthetic data generation for data augmentation

---

Model Architecture Comparison

Category	Model	Description
DL	BiGRU	Captures bidirectional sequential dependencies
DL	BiLSTM	Learns long-term temporal relationships
DL	BiGRU-BiLSTM	Hybrid model combining both architectures
ML	Traditional Models	Baseline classifiers
Ensemble	Ensemble Methods	Combined model predictions

---

Methodology Pipeline

Step	Process	Description
1	Data Collection	Maternal and fetal datasets
2	Preprocessing	Cleaning and normalization
3	Data Balancing	SMOTE and cGAN application
4	Feature Engineering	Feature selection and transformation
5	Model Training	ML, DL, hybrid, and ensemble models
6	Hyperparameter Tuning	Optimization using multiple optimizers
7	Evaluation	Performance and generalization testing

---

Evaluation Metrics

Metric	Purpose
ROC Curve	Classification performance analysis
Confusion Matrix	Prediction distribution evaluation
Accuracy	Overall correctness
Loss Curve	Training convergence

---

Validation Techniques

Technique	Description
Cross-Validation	Multi-fold validation for robustness
Cross-Dataset Analysis	Generalization across datasets
Hyperparameter Tuning	Model optimization

---

Results

Model	Dataset	Accuracy
BiGRU-BiLSTM	Maternal Health	96.21%
BiGRU-BiLSTM	Fetal Health	97.38%
BiGRU-BiLSTM	Merged Dataset	>85%

---

Model Performance Summary

Aspect	Observation
Best Model	BiGRU-BiLSTM
Performance	High accuracy across datasets
Generalization	Strong across cross-data evaluation
Stability	Consistent with validation techniques

---

Explainability Analysis

Method	Purpose
SHAP	Identifies key feature contributions
LIME	Provides local interpretability of predictions

---

Comparative Analysis

Technique	Outcome
Traditional ML	Lower performance
Standalone DL	Moderate performance
Ensemble Learning	Competitive results
Proposed Hybrid Model	Best overall performance

---

Key Contributions

• Hybrid BiGRU-BiLSTM architecture for improved prediction
• Integration of SMOTE and cGAN for data augmentation
• Use of explainable AI techniques (SHAP and LIME)
• Multi-dataset evaluation (maternal, fetal, merged)
• Strong generalization and interpretability

---

Project Structure

Directory/File	Description
data/	Maternal, fetal, and merged datasets
preprocessing/	Data cleaning and augmentation scripts
models/	Model architectures and trained weights
experiments/	Training and evaluation pipelines
results/	Metrics, plots, and analysis outputs
explainability/	SHAP and LIME analysis scripts
README.md	Project documentation

---

Ethical Considerations

Concern	Description
Data Privacy	Sensitive health data must be protected
Bias	Model performance may vary across populations
Clinical Use	Not intended for direct medical diagnosis

---

Disclaimer

This project is intended strictly for research and academic purposes. It is not a substitute for professional medical advice, diagnosis, or treatment.

---

Keywords

Domain	Terms
Application	Pregnancy Risk Prediction, Maternal Health, Fetal Health
Techniques	Machine Learning, Deep Learning, Hybrid Models
Methods	SMOTE, cGAN, Ensemble Learning
Explainability	SHAP, LIME

Citations

@INPROCEEDINGS{10939599, author={Muntaha, Sidratul and Dewanjee, Swarup}, booktitle={2024 International Conference on Innovations in Science, Engineering and Technology (ICISET)}, title={Integrating XAI with Hybrid BiGRU-BiLSTM Model for Comprehensive Maternal-Fetal Health Risk Monitoring}, year={2024}, volume={}, number={}, pages={1-6}, keywords={Pregnancy;Adaptation models;Analytical models;Solid modeling;Accuracy;Data models;Complexity theory;Monitoring;Testing;Synthetic data;Maternal Fetal Health;Pregnancy Risks;cGAN;Explainable AI;Hybrid BiGRU-BiLSTM;LIME;SHAP}, doi={10.1109/ICISET62123.2024.10939599}}