They used a machine learning model that combines blockchain and federated learning, specifically blockchain federated Gaussian multi-agent Q-encoder neural networks (BFGMAQENN), to detect cyberattacks. Additionally, they employed an optimization algorithm, whale swarm binary wolf optimization (WSBWO), to enhance the 6G network. The method was tested for high detection accuracy, data integrity, scalability, and network efficiency.
Background
While fifth-generation (5G) deployment continues, research is already focused on developing 6G networks to support future advanced applications. However, 6G's complexity and the proliferation of IoT devices introduce severe security challenges.
Existing IDS often struggle, as signature-based methods cannot identify novel zero-day attacks, and anomaly-based approaches suffer from high false positives. Traditional machine learning models also tend to operate in isolation, making them vulnerable to sophisticated, distributed threats like data poisoning.
This paper addressed these gaps by proposing a novel, secure IDS for 6G networks. It integrated blockchain with federated learning to enable collaborative, privacy-preserving model training across devices, enhancing detection of new attacks while mitigating data poisoning risks. The system was further optimized to improve overall network efficiency and security.
The Proposed Model
The proposed model was deployed across a diverse 6G infrastructure, including Internet of Things (IoT), mobile, and satellite networks. It utilized a blockchain-based federated learning approach, where high-performance access gateways train local models on their own data without sharing it, thus preserving privacy.
These local models were then aggregated into a global model on a central parameter server. The BFGMAQENN enhanced this federated learning process by incorporating a divergence estimation technique (using Kullback–Leibler divergence) to weigh each client's contribution, ensuring that data distributions closer to the global model have a greater influence.
This created a more robust and accurate system resistant to data poisoning and other distributed attacks. The Q-encoder neural network architecture itself uses efficient deconvolution layers to intelligently upsample and enrich feature maps for better anomaly detection.
Following the training of the BFGMAQENN model, the WSBWO algorithm was applied to optimize the 6G network parameters further. This hybrid optimizer combined the principles of the whale swarm algorithm, which simulates how whales communicate and move towards the "better and nearest" solution, with the grey wolf optimizer, which mimics the social hierarchy and hunting tactics of grey wolves.
The algorithm was adapted for binary optimization, allowing it to effectively select the most relevant features and fine-tune the network for peak performance, balancing high detection accuracy with a minimal number of selected features.
Robust Performance, High Accuracy, and Resilience Across Threats and Distributed Environments
The proposed model was evaluated using multiple benchmark datasets, including CIDDS-001 (Coburg Intrusion Detection Data Sets), KDD98, CAIDA, and KYOTO. Its performance was compared against established methods, specifically the LA-HLRW and SSHA baseline models.
Across all datasets, the model consistently delivered state-of-the-art results. On the KYOTO dataset, it achieved a peak detection accuracy of 97 %, significantly outperforming the baseline approaches. It also demonstrated strong performance in key areas - reporting 94 % data integrity, 93 % scalability, and 98 % network efficiency, all while maintaining a relatively low communication overhead of 60 %.
A major strength of the model lies in its robustness against advanced security threats. It withstood backdoor, data poisoning, and model evasion attacks, showing notable improvements in classification accuracy across all datasets when the defense mechanisms were activated.
The system also handled data heterogeneity effectively. Detection accuracy remained high even as heterogeneity levels increased, dropping only slightly from 97 % at low heterogeneity to 90 % at high, highlighting the model’s adaptability in distributed environments.
To improve transparency, interpretability analyses were conducted using SHAP (Shapley Additive Explanations) and LIME (Local Interpretable Model-Agnostic Explanations). These tools identified critical features influencing detection outcomes, such as packet counts, protocol anomalies, and unusual port activity.
Additionally, the analysis of communication overhead showed a favorable trade-off. By tuning the model update frequency, users could prioritize either peak detection accuracy (97 % at five communication rounds) or enhanced communication efficiency (up to 85 % overhead reduction at 80 rounds), offering operational flexibility.
Finally, the integration of blockchain-supported federated learning proved beneficial. The system maintained stability and convergence even under conditions of packet loss during weight transmission, confirming its resilience in real-world network environments.
Conclusion
This work marks a critical step toward securing 6G networks through intelligent, decentralized defense mechanisms. By integrating blockchain with federated learning and optimizing through WSBWO, the proposed BFGMAQENN model not only addresses long-standing challenges in intrusion detection but also introduces a scalable framework for real-world deployment.
What sets this approach apart is its ability to operate effectively in adversarial, resource-constrained, and highly heterogeneous environments, which are expected to define the landscape of next-generation wireless systems. The model’s resilience to sophisticated attacks and its tunable performance characteristics position it as a practical solution for dynamic, high-stakes scenarios where centralized defenses fall short.
Future work must move beyond technical refinement and explore operational integration. This includes real-time deployment across edge infrastructure, continuous learning from evolving threats, and improved handling of underrepresented attack classes.
Ultimately, the goal is not just better detection, but a shift toward autonomous, adaptive, and trust-aware network security - foundational to the future of 6G and beyond.
Journal Reference
Chinnasamy, P., Yarramsetti, S., Ayyasamy, R. K., Rajesh, E., Vijayasaro, V., Pandey, D., Pandey, B. K., & Lelish, M. E. (2025). AI-Driven Intrusion Detection and Prevention Systems to Safeguard 6G Networks from Cyber Threats. Scientific Reports, 15, 37901. https://www.nature.com/articles/s41598-025-21648-5
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.