FedSec-CPS: Federated Learning Based Security Framework for Security of Cyber-Physical Systems

The proliferation of cyber-physical systems (CPS) has posed significant challenges in ensuring their security due to their distributed nature and resource constraints. Machine learning (ML) techniques have been utilized to detect known and zero-day attacks. However, traditional ML algorithms may exceed the computational requirements of CPS, leading to poor performance and efficiency. A potential strategy to enhance performance and efficiency of ML algorithms on constrained devices while also providing a more secure environment for data sharing is the use of Federated Learning (FL). This paper proposes a methodology for computationally efficient cyber security in constrained CPS using FL. The methodology utilizes Horizontal Federated Learning (HFL), and Vertical Federated Learning (VFL) with Random Forest (RF) as the base classifier. We evaluate the proposed methodology on TON-IoT (network traffic) dataset. Performance metrics of our model using RF with FL architecture ranged from 74% to 100%, while computation time was low. FL-based methodology achieved better performance and efficiency while preserving the privacy of CPS data. Moreover, this paper emphasizes the potential of FL to address the security and performance challenges of traditional ML algorithms in constrained CPS.