ORCID
- Muhammad Asad: 0000-0003-0036-1714
Document Type
Article
Abstract
Federated intrusion detection enables collaborative model training across distributed Internet of Things (IoT) environments without centralizing raw telemetry. However, its effectiveness is challenged by heterogeneous local data and the participation of malicious or unreliable clients. Existing federated intrusion detection approaches generally address multi-source feature representation and adversarially robust aggregation as separate problems. This paper presents FedFusion, a robust federated intrusion detection framework that combines attention-based multi-source data fusion at edge nodes with trust-aware aggregation at the federated coordinator. Modality-specific encoders extract representations from heterogeneous telemetry sources, while an attention mechanism learns their relative importance for intrusion detection. During federated training, the reliability of each client is estimated from the consistency of its model update, and the resulting trust score is incorporated into the aggregation process to reduce the influence of anomalous or poisoned updates. FedFusion is evaluated on the Edge-IIoTset and IoT-23 datasets under IID, non-IID, and malicious-client settings using label-flipping and Gaussian-noise model-poisoning attacks. The proposed framework achieves F1-scores of 96.34% and 92.10% on Edge-IIoTset and IoT-23, respectively, and demonstrates lower performance degradation than the evaluated federated intrusion detection baselines when the proportion of malicious clients increases to 50%. These results demonstrate that combining multi-source data fusion with trust-aware aggregation can improve the detection performance, robustness, and training stability of federated intrusion detection in heterogeneous and adversarial IoT environments.
DOI Link
Publication Date
2026-11-02
Publication Title
IEEE Access
ISSN
2169-3536
Deposit Date
2026-11-04
Keywords
Federated Learning, intrusion detection, IoT Security, Trust-Aware Aggregation, Adversarial Robustness
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 International License.
First Page
1
Last Page
1
Recommended Citation
Banbhrani, S., Wei, C., & Asad, M. (2026) 'A Trust-Aware Federated Learning Framework with Multi-Source Data Fusion for Robust IoT Intrusion Detection', IEEE Access, , pp. 1-1. Available at: 10.1109/ACCESS.2026.3749183
