International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 05 | May 2025
p-ISSN: 2395-0072
www.irjet.net
Federated Learning-Based Intrusion Detection Systems for Transportation IoT: A Comparative Study on Efficiency, Privacy, and Scalability Farzana Anjum G1, Divya Rani2, Aparna Singh3, Dr. Nirmala S4 1,2,3Student, Dept. of Computer Science Engineering, AMC Engineering College, Karnataka, India
, Dept. of Computer Science Engineering, AMC Engineering College, Karnataka, India ---------------------------------------------------------------------***--------------------------------------------------------------------4Professor
Abstract - With the widespread integration of Internet of
collaboratively develop models without exchanging raw data. This approach maintains user privacy while enhancing adaptability and responsiveness across heterogeneous networks. In the context of transportation IoT, FL-based IDS architectures have shown potential in detecting complex and evolving attack patterns while operating efficiently on edge devices.
Things (IoT) devices in transportation systems, maintaining data confidentiality and system reliability has become increasingly critical. Traditional intrusion detection systems (IDS), typically built on conventional machine learning models, face challenges including data privacy risks, high communication overhead, and reduced responsiveness in real-time scenarios. Federated Learning (FL) offers a decentralized approach that enables local model training across distributed devices, preserving data privacy while leveraging collective intelligence. This paper provides a comparative analysis of four FL-based IDS frameworks developed for transportation IoT environments. Each study introduces unique methodologies—ranging from lightweight fine-tuning techniques to feature optimization strategies and real-time deployments on edge hardware. The analysis focuses on evaluating detection accuracy, system efficiency, scalability, and deployment feasibility. The comparative findings highlight critical design considerations and outline potential pathways for developing effective, privacypreserving IDS solutions tailored for the transportation sector.
This study presents a comparative review of four contemporary FL-IDS models that aim to strengthen cybersecurity in transportation networks. These comprise systems engineered for resource-constrained vehicular nodes, models optimized through hybrid server-edge learning, and frameworks integrating deep learning integrated with feature selection algorithms. By analysing each approach in terms of accuracy, deployment strategy, and system requirements, this paper offers significant understanding of the current landscape of FL-IDS and identifies potential directions for future research.
2. LITERATURE SURVEY The implementation of Federated Learning (FL) with Intrusion Detection Systems (IDS) has emerged as a significant advancement in the cybersecurity landscape of Transportation IoT (T-IoT). Several recent studies propose FL-based frameworks to mitigate evolving cyber threats while safeguarding data privacy and reducing centralized dependencies.
Key Words: Federated Learning, Intrusion Detection System (IDS), Transportation IoT, Edge Computing, Cybersecurity, Privacy Preservation, Distributed Learning
1.INTRODUCTION In recent years, smart transportation systems have rapidly adopted Internet of Things (IoT) technologies, enabling advanced functionalities such as automated driving, realtime traffic monitoring, and vehicle-to-everything (V2X) communication. While these developments enhance operational efficiency and safety, they also expose systems to a wide array of cyber threats. Traditional IDS frameworks often rely on centralized data collection and model training, which can lead to significant drawbacks, including compromised privacy, communication bottlenecks, and limited scalability in dynamic vehicular environments.
In the study by Bhavsar et al. [4], an FL-IDS system optimized for vehicular environments is introduced. This model employs a blend of logistic regression and convolutional neural networks (CNNs) and is deployed on real-time edge platforms such as Raspberry Pi and Jetson Xavier. By utilizing embedded devices for local model training and a central aggregator for global model refinement, this architecture successfully maintains data privacy while maintaining high detection accuracy. Experiments conducted using the NSL-KDD and Car-Hacking datasets demonstrated a performance gain over centralized IDS models, achieving accuracies of up to 99%.
In order to overcome these obstacles, researchers have explored Federated Learning (FL), a distributed machine learning framework that allows distributed devices to
Lazzarini et al. [2] propose a lightweight FL-based IDS framework for IoT systems that utilizes a shallow artificial
© 2025, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 470