Collaboration-based Privacy-Preserving Firewall for Home Networks

Rahimi Sadegh, Yasamin; Shirali-Shahreza, Sajad

doi:10.22042/isecure.2026.256933

Collaboration-based Privacy-Preserving Firewall for Home Networks

Articles in Press

Document Type : Research Article

Authors

Yasamin Rahimi Sadegh

Sajad Shirali-Shahreza

Department of Computer Engineering, Amirkabir University of Technology, Tehran, Iran

10.22042/isecure.2026.256933

Abstract

The rise of IoT devices has increased security risks, requiring faster and more effective defenses. Although home gateways are the first line of defense, they struggle against advanced threats. Current learning-based detection methods offer speed, but face scalability and privacy issues in distributed IoT environments. To address this, our research proposes a collaborative firewall using federated learning. It enables attack detection at the gateway while maintaining raw training records on participating clients. The firewall employs federated tree models that exchange training summaries with an aggregator rather than centrally collecting raw gateway traffic. On Edge-IIoTset dataset, our feature selection reduces the number of input features from 61 to 22, cutting training time by 75% and per-packet inference time by more than 65%, while the accuracy slightly reduced from 93.6% to 92.53%. When the federated training was used, the accuracy slightly increased to 93.8%. This shows that collaborative training did not reduce detection accuracy in our setting. Our framework combines local attack detection, collaborative training, and reduced feature and execution costs for smart-home security.

Keywords

Federated Learning

Firewall

Smart Home

IoT Security