Privacy-Preserving Explainable Network Anomaly Detection in Distributed Edge Environments with Federated Ensemble Learning
Journal of Innovative Research and Technology · 25.06.2026 · 3 (1) · 200-209
DOI: 10.66857/jirt.v3i1.46 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Innovative Research and Technology, online erschienen am 25.06.2026. Verfasst von Matthew Clarke und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Distributed edge networks generate large volumes of heterogeneous traffic data, but privacy and data-sharing restrictions often prevent centralized model training for network anomaly detection. Federated learning provides a potential solution, yet most federated intrusion detection methods still lack interpretability and are vulnerable to unstable performance across heterogeneous clients. This study proposes a privacy-preserving explainable anomaly detection framework based on federated ensemble learning for distributed edge security environments. The proposed method trains local tree-based classifiers, including Random Forest and LightGBM, on multiple edge clients and aggregates model-level predictions through a federated soft-voting mechanism. To improve transparency, local and global SHAP explanations are generated to identify consistent attack-related features across clients without exposing raw traffic records. The framework is tested on partitioned CIC-ToN-IoT and Bot-IoT datasets, with 10 simulated edge clients and more than 3.1 million labeled network flow samples. Non-IID data partitions are constructed to reflect realistic deployment conditions, where each client contains different attack distributions and traffic volumes. The proposed model achieves an average accuracy of 97.64%, macro-F1 of 96.39%, and false positive rate of 1.92% across all clients. Compared with centralized LightGBM, the federated ensemble model reduces raw data exposure while maintaining 98.7% of the centralized model’s detection performance. Compared with FedAvg-based neural network baselines, it improves macro-F1 by 3.28% under non-IID traffic settings. SHAP interpretation shows that packet rate, byte asymmetry, connection duration, and destination port concentration remain stable indicators across distributed clients. The results indicate that explainable federated ensemble learning is suitable for privacy-aware anomaly detection in edge and embedded network environments
Zitieren
APA
Clarke, M., Thompson, E., & Foster, D. (2026). Privacy-Preserving Explainable Network Anomaly Detection in Distributed Edge Environments with Federated Ensemble Learning. Journal of Innovative Research and Technology, 3(1), 200-209. https://doi.org/10.66857/jirt.v3i1.46
Vancouver
Clarke M, Thompson E, Foster D. Privacy-Preserving Explainable Network Anomaly Detection in Distributed Edge Environments with Federated Ensemble Learning. Journal of Innovative Research and Technology. 2026;3(1):200-209. doi:10.66857/jirt.v3i1.46
BibTeX
@article{Clarke2026,
author = {Clarke, Matthew and Thompson, Emily and Foster, Daniel},
title = {Privacy-Preserving Explainable Network Anomaly Detection in Distributed Edge Environments with Federated Ensemble Learning},
journal = {Journal of Innovative Research and Technology},
year = {2026},
volume = {3},
number = {1},
pages = {200--209},
doi = {10.66857/jirt.v3i1.46}
}
EUVIMED