Table 1.
Summary of key related studies in federated, privacy-preserving, and multimodal healthcare learning.
Fig 1.
System architecture overview: IoT and EMR layers with encoders, detectors, fusion head, and final prediction path.
Fig 2.
Preprocessing and learning pipeline: data cleaning, encoding, outlier filtering, dimensionality reduction, class balancing, and stratified splits.
Table 2.
Architectural details of IoT branch, EMR branch, and Fusion head.
Table 3.
Performance comparison on IoT dataset.
Table 4.
Performance comparison on EMR dataset.
Table 5.
Performance comparison on combined IoT and EMR dataset.
Table 6.
Resilience under client poisoning (combined dataset).
Table 7.
Impact of communication constraints (combined dataset).
Table 8.
Impact of Privacy Mechanisms on Model Performance (combined dataset).
Table 9.
Detection latency comparison on IoT dataset. Lower is better.
Table 10.
Ablation study on combined dataset. Each row removes a component from the proposed pipeline.
Fig 3.
Comparative evaluation of the proposed pipeline against baselines.
(a) AUPRC highlights superior detection quality. (b) Robust accuracy is sustained even with malicious clients. (c) Lower and more stable IoT detection latency ensures timely alerts.
Fig 4.
Training dynamics of the proposed pipeline.
(a) Loss curves show smooth convergence with small generalization gap. (b) Accuracy curves indicate stable generalization and controlled overfitting.
Fig 5.
Evolution of ranking-based performance metrics over epochs.
(a) AUPRC reflects precision–recall trade-offs under class imbalance. (b) AUROC illustrates class separability improvements.
Fig 6.
Training efficiency of the proposed pipeline.
(a) Adaptive learning rate schedule accelerates convergence while maintaining stability. (b) Stable epoch times confirm scalability of the distributed implementation.
Fig 7.
Federated training dynamics of the proposed pipeline.
(a) Validation AUPRC convergence demonstrates stable distributed optimization. (b) Privacy budget accumulation reflects the trade-off between utility and differential privacy guarantees.
Fig 8.
Robustness and reliability of the proposed pipeline.
(a) Defense mechanisms sustain accuracy under poisoning. (b) Probability calibration improves steadily, enabling reliable risk-aware decision-making.