Securing the Edge: An AI-Driven Federated Unlearning Framework for Cybersecurity and IoT Forensics
Year: 2026
Type: article
Abstract: The rapid proliferation of Internet of Things (IoT) devices has expanded the digital attack surface, complicating cybersecurity monitoring and digital forensics. While Federated Learning (FL) offers a privacy-preserving alternative to centralized machine learning by training models across distributed devices without sharing raw data, it struggles with "unlearning" specific contributions. Removing the influence of poisoned updates, compromised nodes, or users exercising their "right to be forgotten" is traditionally computationally expensive. To address this, it is proposed to use the Efficient Federated Unlearning (EFU) framework for privacy-preserving IoT forensic intelligence. EFU enables the selective removal of device contributions from a global model without full retraining. The architecture integrates an adaptive unlearning controller with influence function analysis to estimate client updates and employs knowledge distillation to maintain model accuracy on resource-constrained devices. Experimental evaluations using IoT security datasets show that EFU significantly reduces computational overhead and convergence time compared to retraining-based methods while maintaining high detection accuracy and resilience against backdoor attacks. This scalable protocol strengthens regulatory compliance and reliability in smart cities and industrial IoT (IIoT) environments.
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7160051817