Chronic obstructive pulmonary disease (COPD) exacerbations are difficult to anticipate because deterioration develops across heterogeneous respiratory, physiological, behavioral, environmental, and clinical signals. This study presents FedSqueeze-HGRUViTNet, a lightweight federated deep learning model for COPD exacerbation-risk prediction using multimodal wearable and harmonized clinical information. The model integrates depthwise separable convolution for local health-index variation, a health-gated residual gated recurrent unit for patient-specific temporal deterioration, and a squeeze-enhanced Health Vision Transformer for broader multimodal interactions. Personalized baseline deviation, explicit missingness indicators, squeeze-based recalibration, adaptive feature fusion, and weighted federated averaging are incorporated to support distributed learning without centralizing raw patient records. The model was evaluated on the TOLIFE COPD Health Indices Dataset and a harmonized Medical Information Mart for Intensive Care IV (MIMIC-IV) COPD cohort using patient-level 70:10:20 training, validation, and testing partitions. Each experiment was repeated across five random seeds using fixed patient partitions. On TOLIFE, FedSqueeze-HGRUViTNet achieved 97.30% accuracy and an area under the receiver operating characteristic curve (AUC) of 0.985 for binary exacerbation-risk prediction, together with 96.72% accuracy for three-class risk stratification. On the complementary MIMIC-IV electronic health record-based evaluation, the model achieved 96.62% accuracy and an AUC of 0.978. Robustness experiments under missing observations, Gaussian-noise perturbation, monitoring-window variation, and non-independent and non-identically distributed client conditions showed limited performance degradation, with accuracy remaining above 95.6% across the evaluated settings. Inference required 5.18 ms per sample in the evaluated computing environment. These results indicate that FedSqueeze-HGRUViTNet provides an accurate and computationally efficient federated approach for distributed COPD exacerbation-risk assessment across heterogeneous multimodal health information.
Alharbey, R. (2026). Lightweight Federated Deep Learning for COPD Exacerbation Risk Prediction Using Multimodal Ubiquitous Health Sensing. International Journal of Advanced Computer Science and Applications, 17(10). https://doi.org/10.14569/IJACSA.2026.01710113
Alharbey, Riad. "Lightweight Federated Deep Learning for COPD Exacerbation Risk Prediction Using Multimodal Ubiquitous Health Sensing." International Journal of Advanced Computer Science and Applications, vol. 17, no. 10, 2026, https://doi.org/10.14569/IJACSA.2026.01710113.
@article{Alharbey2026,
title = {Lightweight Federated Deep Learning for COPD Exacerbation Risk Prediction Using Multimodal Ubiquitous Health Sensing},
journal = {International Journal of Advanced Computer Science and Applications},
volume = {17},
number = {10},
year = {2026},
publisher = {The Science and Information Organization},
author = {Riad Alharbey},
doi = {10.14569/IJACSA.2026.01710113},
url = {https://doi.org/10.14569/IJACSA.2026.01710113}
}
Open Access — licensed under a
Creative Commons Attribution 4.0 International License.
Unrestricted use, distribution, and reproduction in any medium, even commercially, as long as the original work is properly cited.