Related Experiment Videos
Adaptive Multi-Model Hierarchical Federated Learning for Robust IoT Intrusion Detection
Shahid Latif1, Djamel Djenouri1
1School of Computing and Creative Technologies, University of the West of England, Bristol BS16 1QY, UK.
Sensors (Basel, Switzerland)
|May 27, 2026
Summary
This study introduces an Adaptive Multi-Model Hierarchical Federated Learning (AMM-HFL) framework to enhance cybersecurity for the Internet of Things (IoT). AMM-HFL improves intrusion detection accuracy, especially with diverse data, by using multiple adaptive models.
Area of Science:
- Cybersecurity
- Machine Learning
- Distributed Systems
Background:
- The Internet of Things (IoT) presents significant cybersecurity challenges due to its distributed and heterogeneous nature.
- Traditional intrusion detection systems and federated learning (FL) struggle with non-IID data and adversarial attacks in IoT environments.
- Centralized aggregation methods in FL are insufficient for robust IoT security.
Purpose of the Study:
- To propose an Adaptive Multi-Model Hierarchical Federated Learning (AMM-HFL) framework for robust IoT intrusion detection.
- To address the limitations of traditional FL in handling extreme non-IID data and adversarial conditions.
- To enhance the accuracy and adaptability of intrusion detection in complex IoT ecosystems.
Main Methods:
- Developed a three-tier AMM-HFL framework (client, edge, cloud) integrating similarity-aware clustering, multi-model aggregation, and dynamic client-side model selection.
- Maintained multiple global models for adaptive personalization and better representation of heterogeneous data distributions.
- Implemented edge-level clustering of model updates to isolate anomalies and cloud-level meta-aggregation for refining diverse models.
Main Results:
- Achieved high detection accuracy on the IDSIoT2024 dataset: up to 97.54% under IID and 97.52% under non-IID conditions.
- Demonstrated the framework's robustness against extreme non-IID data and adversarial scenarios.
- Maintained low computational and cryptographic overhead compared to traditional methods.
Conclusions:
- The AMM-HFL framework offers a significant advancement in IoT intrusion detection by effectively managing data heterogeneity and adversarial threats.
- The multi-model and hierarchical approach provides superior adaptability and accuracy compared to single-model FL methods.
- AMM-HFL presents a scalable and efficient solution for securing the rapidly growing landscape of IoT devices.
