Lightweight Hybrid Ensembles for Antenatal Malnutrition Risk Prediction in Women | IJCT Volume 13 – Issue 1 | IJCT-V13I1P23
International Journal of Computer Techniques
ISSN 2394-2231
Author
V. Saravanakumar
Abstract
Antenatal malnutrition poses severe risks to maternal and fetal health, yet early identification in primary care is often hindered by resource limitations. This study presents a lightweight hybrid ensemble framework for the automated prediction of malnutrition risk in pregnant women. Leveraging data from Point of Care forms, EMRs, and screening logs, the system integrates clinical biomarkers (e.g., haemoglobin, parity), anthropometrics (MUAC, BMI), nutritional adherence, and socio demographic factors to derive a binary risk flag. The methodology addresses real world data sparsity through deterministic preprocessing, utilizing median imputation with missingness indicators and target encoding. The core architecture employs a stacked ensemble strategy, combining the non-linear strengths of tree based algorithms with the stability of linear models. This hybrid approach optimizes predictive accuracy while maintaining a low computational footprint suitable for deployment in resource constrained settings. To ensure clinical utility, the model incorporates probability calibration via isotonic regression. Crucially, the system embeds an explainability layer using SHAP (SHapley Additive exPlanations) to provide local, instance level reasoning, fostering trust among healthcare providers. Designed for sustainable operations, the architecture features robust governance protocols, including performance monitoring and feature drift detection. This approach demonstrates that hybrid ensembles can effectively bridge the gap between complex predictive analytics and practical clinical application, enabling timely, data driven interventions for risk pregnancies.
Keywords
Malnutrition, Multimodal, EMR, LightGBM, SHAP, Ensembles.
Conclusion
This study successfully introduced and validated a novel, operationally ready Lightweight Hybrid Ensemble Stacking framework for the accurate and transparent prediction of antenatal malnutrition risk in women attending decentralized primary care clinics [Sinha & Devi, 2021]. Addressing the dual challenge of high performance modelling and constrained computing resources, our methodology combined the high discriminatory power of non-linear LightGBM with the stability and probabilistic reliability of Logistic Regression via a simple meta-learner. The empirical results conclusively demonstrated the system’s superior capability in risk stratification, achieving a mean AUC ROC of $0.92$ and an optimized F1 Score of $0.80$, significantly surpassing all tested single learner baselines [Wang et al., 2020]. Crucially, the final output was meticulously calibrated using Isotonic Regression, yielding an extremely low Brier Score of $0.04$, ensuring that the predicted risk probabilities are statistically reliable for clinical action [Zadrozny & Elkan, 2001]. The integrated SHAP based explainability framework, utilizing fast approximations, provides essential transparency, validating the model’s reliance on clinically relevant features such as MUAC and Haemoglobin and delivering immediate, patient specific rationales to clinicians at the point of care [Lundberg & Lee, 2017].The fundamental contribution of this work lies in successfully marrying algorithmic sophistication with deployment pragmatism. By designing the system for low compute inference and integrating a robust governance lifecycle that includes continuous feature drift detection and a structured clinician feedback loop, we have provided a solution that is not only accurate but also sustainable and ethically accountable over the long term. This scalable architecture empowers healthcare providers to transition from reactive treatment to proactive, data driven intervention, ultimately promising improved maternal and infant health outcomes in resource constrained environments. Future work should focus on prospective validation of the system in a real world, multi-site deployment, evaluating its longitudinal impact on clinical referral rates and patient outcomes. Furthermore, integration with decentralized mobile health platforms should be explored to maximize accessibility for community health workers.
References
[1].R. Sinha and A. Devi, “Global prevalence and impact of antenatal malnutrition on infant mortality,” The Lancet Global Health, vol. 9, no. 4, pp. e447–e456, 2021.
[2].World Health Organization, Maternal Nutrition and Health: A Global Priority. Geneva: WHO, 2020.
[3].Jha, R. Kumar, and V. Sharma, “The intergenerational cycle of poverty and maternal health outcomes,” International Journal of Social Determinants of Health, vol. 12, no. 3, pp. 321–335, 2017.
[4].W. Cai and C. Lin, “Limitations of single metric screening tools for nutritional risk,” Public Health Nutrition, vol. 22, no. 10, pp. 1801–1810, 2019.
[5].K. Tiwari, “Multi-factorial assessment of nutritional deficits in rural pregnant populations,” Global Health Action, vol. 15, no. 1, p. 2098765, 2022.
[6].J. Garcia, A. Rodriguez, and S. Lee, “Data fragmentation and logistical lags in community antenatal care,” Community Health Journal, vol. 54, no. 4, pp. 589–601, 2018.
[7].L. Wang et al., “Predicting pre-term birth using XGBoost and multi-source electronic health records,” Journal of Biomedical Informatics, vol. 108, p. 103498, 2020.
[8].Deep Health Consortium, “Bias and performance variance in large-scale clinical deep learning models,” Nature Digital Medicine, vol. 4, no. 1, pp. 45–58, 2023.
[9].M. Ribeiro and D. Silva, “The black box problem in healthcare AI: Challenges for adoption and trust,” Health Policy and Technology, vol. 8, no. 3, pp. 256–263, 2019.
[10].Gartner, Hype Cycle for AI in Healthcare 2024. Technical Report, 2024.
[11].L. Breiman, “Random Forests,” Machine Learning, vol. 45, no. 1, pp. 5–32, 2001.
[12].T. G. Dietterich, “Ensemble Methods in Machine Learning,” in Proceedings of the First International Workshop on Multiple Classifier Systems, Springer, 2000.
[13].P. Kaur and N. Singh, “Simple ensemble methods for mitigating data imbalance in maternal health prediction,”Science, vol. 20, no. 4, pp. 887–905, 2022.
[14].S. M. Lundberg and S. Lee, “A unified approach to interpreting model predictions,” in Advances in Neural Information Processing Systems (NeurIPS), pp. 4765–4774, 2017.
[15].R. Acosta, M. Sanchez, and L. Perez, “The necessity of missingness indicators in clinical data imputation,” Journal of Health Informatics, vol. 5, no. 3, pp. 112–125, 2018.
[16].P. Gessner, “The predictive power of missingness: Feature engineering for sparsity,” IEEE Transactions on Knowledge and Data Engineering, vol. 31, no. 8, pp. 1545–1557, 2019.
[17].X. Tong, “The transformative role of feature engineering in deep clinical prediction,” IEEE Transactions on Biomedical Engineering, vol. 67, no. 5, pp. 1432–1440, 2020.
[18].J. Schmidhuber, Continuous Learning and Governance Frameworks for AI Deployment. Technical Report, AI Research Group, 2023.
[19].Smith and B. Jones, “Operationalizing transparent AI: A guide for clinical governance,” Journal of Medical Systems, vol. 48, no. 1, pp. 1–12, 2024.
[20].M. Al Mashrafi, N. Al Busaidi, and S. Al Hinai, “Operationalizing clinical machine learning in low-resource primary care environments,” in Proceedings of the International Conference on Health Technology, pp. 202–210, 2024.
[21].C. Turner, “Model quantization and sparse representations for edge computing in healthcare,” Sensors, vol. 21, no. 18, p. 6160, 2021.
[22].Maheshwari, “Monitoring data and concept drift in live machine learning systems,” ACM Computing Surveys, vol. 54, no. 5, pp. 1–38, 2021.
[23].Q. Zhai, “Detecting model decay and concept shift in continuous clinical deployment,” International Journal of Medical Informatics, vol. 174, p. 105151, 2023.
[24].S. Patel et al., “Detection of feature distribution shifts using Jensen-Shannon divergence in clinical trials,” Statistical Methods in Medical Research, vol. 31, no. 9, pp. 1735–1750, 2022.
[25].S. Verma and J. Rubin, “Fairness definitions explained,” in Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society, pp. 23–34, 2018.
[26].D. Hayes and A. Nolan, “Designing accountable AI: The role of the human in the loop and feedback mechanisms,” Ethics and Information Technology, vol. 25, no. 2, pp. 1–15, 2023.
[27].University of Oslo, HAPI FHIR: Fast Healthcare Interoperability Resources Implementation. Software Documentation, 2020.
[28].T. Robinson, Building High Performance REST APIs with Python. O’Reilly Media, 2020.
[29].B. Zadrozny and C. Elkan, “Obtaining calibrated probability estimates from decision trees and naive Bayes,” in Proceedings of the Eighteenth International Conference on Machine Learning (ICML), pp. 609–616, 2001.
S. Kilicarslan et al., “The advantage of hybrid ensemble stacking in clinical risk assessment,” IEEE Journal of Biomedical and Health Informatics, vol. 25, no. 6, pp. 2100–2110, 2021.
How to Cite This Paper
V. Saravanakumar (2025). Lightweight Hybrid Ensembles for Antenatal Malnutrition Risk Prediction in Women . International Journal of Computer Techniques, 12(6). ISSN: 2394-2231.
