Federated Learning-based Intrusion Detection System (IDS) with Explainability of Classification Decisions for Internet of Vehicles (IoV) Networks
Open Access DepositedThe Internet of Vehicles (IoV) represents a significant evolution in the transportation sector. It enhances intelligent transportation systems (ITS) by utilizing advanced sensory data processing and computational resources at the edge. Intelligent Connected Vehicles (ICVs) offer innovative applications that improve road safety, optimize traffic flow, and enable autonomous driving and intelligent navigation. IoV facilitates real-time and interactive communication between vehicles, infrastructure, and users. Wireless communication shares critical data such as traffic conditions and road hazards. Exchanging data enhances driver safety, decision-making and emergency response. Issues attributed to inadequate cybersecurity, privacy, communication protocols, and compliance to regulatory framework are major challenges that IoV systems face. Federated Learning (FL) presents a decentralized solution for developing intrusion detection systems (IDS) that protect user privacy and improve security. In this research, we introduce five FL-based IDS specifically designed for IoV networks, focusing on improving detection Accuracy and training efficiency. We evaluate the performance of three key communication protocols—Synchronous FL (SFL), Asynchronous FL (AFL), and Semi-Asynchronous FL (SAFL)—in terms of their impact on training time and Accuracy. By addressing issues such as the straggler effect and model staleness, we propose innovative algorithms incorporating time-based mechanism, federated dropout, model version control, and synchronization to optimize IDS performance. Our approach involves edge servers receiving local model updates from vehicles (clients) and communicating these updates to a central cloud server for global model aggregation. This architecture enhances traditional FL methods, integrating synchronous, asynchronous, and semi-asynchronous communication protocols for IoV-based IDS. We further enhance the interpretability and transparency of the IDS by incorporating Explainable Artificial Intelligence (XAI) techniques and metrics, specifically SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME), to make the system more understandable for human operators.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.