Integrating XAI with Hybrid BiGRU-BiLSTM Model for Comprehensive Maternal-Fetal Health Risk Monitoring

Pregnancy health risk analysis is a critical concern as it impacts the well-being of both mother and the fetus. This task demands frequent assessments and prompt interventions to avoid clinical complexities. Despite medical advancements, challenges persist while monitoring health risk during pregnancies. This study aimed to develop an automated system to mitigate these risk factors by employing Deep Learning (DL) and Machine Learning (ML) approaches along with Hybridization and Ensemble Learning techniques. By leveraging both Maternal and Fetal health datasets, the system intended to aid in taking informed decisions for improving health outcomes. Data quality was ensured through preprocessing, addressing class imbalance and data size limitations with Synthetic Minority Over-sampling Technique (SMOTE) and Conditional Generative Adversarial Network (cGAN). Thorough hyperparameter tuning, including inspection with various optimizers and rigorous evaluation processes was conducted. The generalizability was validated through crossdata analysis and cross-validation. A hybrid BiGRU-BiLSTM model was then proposed for demonstrating superiority with $\mathbf{9 6. 2 1 \%}$ and $\mathbf{9 7. 3 8 \%}$ accuracies on Maternal and Fetal datasets respectively. SHapley Additive exPlanations (SHAP) analysis was conducted to interpret model’s predictions and identify key features from existing datasets to create a merged dataset for adaptability evaluation. The proposed model yielded over $\mathbf{8 5 \%}$ accuracy on this data. Subsequently, Local Interpretable Model-Agnostic Explanations (LIME) analysis was availed to gain deeper insights into the concluded predictions. These findings highlight the proposed model’s potential applicability to enhance maternal and fetal health risk detection.