Federated Learning-based Intrusion Detection Framework for Internet of Things and Edge Computing backed Critical Infrastructure

Modern Critical Infrastructure (CI) sectors including Smart Girds operate on the Internet of Things and edge computing paradigm. With the enormous growth of these sectors, there are emerging and escalating cyber threats. Traditional Machine Mearning (ML) approaches strive to provide a certain level of resilience against cyber threats but at the cost of privacy leading towards a single point of vulnerability. Following an exhaustive analysis of traditional ML algorithms and cyber threats to the CI, this work introduces a privacy-preserving Federated Learning (FL) driven intrusion detection framework to identify cyber threats focusing on the use case of Smart Grids within the CI. This paper firstly implements and compares various traditional ML algorithms such as Support Vector Machine, Random Forest, and Logistic Regression which works on a centeralised dataset. Secondly, an analysis has been carried out using the proposed FL-based approach to further improve security and privacy along with minimising the need for centralised dataset. Experimental results highlight that our traditional RF-based approach and FL-based approach achieve high intrusion detection accuracy. However, FL has more significant advantages in distributed and privacy-sensitive environments, protecting privacy and reducing the need for data centralisation.