Federated Learning Augmented Cybersecurity for SDN-Based Aeronautical Communication Network
Muhammad Ali, Yim Fun Hu, Jian-Ping Li · Electronics · 2025
With the requirements of government data protection regulations and industrial concerns regarding data protection and privacy, the security level required for data privacy and protection has increased. This has led researchers to investigate techniques that can train cybersecurity machine learning (ML) models without sharing personal data. Federated Learning (FL) is a newly developed decentralized and distributed ML mechanism that emphasize privacy. In this technique, a learning algorithm is trained without collecting or exchanging sensitive data from distributed client models running at different locations. With the rapid increase in the number of cybersecurity attacks reported in the aviation industry in the last two decades, strong, dynamic, and effective countermeasures are required to protect the aviation industry and air passengers against such attacks, which can most of the time lead to catastrophic situations. This paper proposes and implements an FL model for identifying cyberattacks on a Software Defined Network (SDN)-based aeronautical communication networks. The machine learning model used in the FL architecture is a Deep Neural Network (DNN) model. The publicly available National Security Laboratory–Knowledge Discovery and Datamining (NSL-KDD) dataset was employed to train and validate the proposed FL model. The simulation results illustrated that the FL-based system can accurately and effectively identify potential cybersecurity attacks and minimize the risk of data and service exposure without degrading model performance. A comparison was also made between the FL and non-FL machine learning models. Preliminary results demonstrated that the FL model outperformed the non-FL machine learning approaches. FL reached an accuracy of 96%, compared to 76% and 83% for NFL.