| Automated Ensemble Multimodal Machine Learning for Healthcare | |
|
MSRDG International Journal of Computer Scientific Technology & Electronics Engineering
© 2026 by MSRDG IJCSTEE Journal
Volume 2 Issue 2
Year of Publication: 2026 |
Paper Download Article ID MSRDG-IJCSTEE-V2I2P103 |
|
Abstract: Healthcare data is inherently multimodal, encompassing structured electronic health records (EHR), radiological images, high-dimensional genomic sequences, unstructured clinical narratives, and continuous physiological signals from wearable devices. Conventional machine learning pipelines typically exploit a single modality, thereby neglecting the complementary information latent in other data streams. We introduce an Automated Ensemble Multimodal Machine Learning (AE-MML) framework that systematically integrates five heterogeneous data modalities. The framework comprises modality-specific preprocessing, deep-feature extraction (Convolutional Neural Networks for imaging, Bidirectional Long Short-Term Memory networks for sequential clinical notes), and automated hyperparameter optimisation via Bayesian search implemented through Optuna. A two-level stacking meta-learner aggregates predictions from five diverse base classifiers: Logistic Regression, Support Vector Machine, Random Forest, XGBoost, and a CNN-BiLSTM hybrid. The proposed AE-MML framework is evaluated on a curated multi-source dataset comprising 166,639 patient records drawn from MIMIC-IV, ChestX-ray14, TCGA, i2b2, and WESAD. It achieves an accuracy of 92.4%, F1-score of 91.6%, and AUC-ROC of 94.2%, outperforming all single-modality baselines by at least 5.3 percentage points in AUC-ROC. An ablation study confirms that each modality contributes independently and cumulatively to overall model performance. The AE-MML framework demonstrates that automated ensemble fusion of multimodal clinical data substantially improves predictive accuracy for patient risk stratification. The modular and extensible architecture supports prospective clinical deployment and enables interpretable decision support. Future work will incorporate federated learning to address data-privacy constraints across hospital networks. |
|
| Keywords: Multimodal machine learning, Automated ensemble learning, Stacking meta-learner, EHR analysis, Medical imaging, Clinical NLP , Patient risk stratification , Healthcare AI | |
