Explainable & Privacy-Preserving Misbehavior Detection in Next-Generation Vehicular Networks
Format
Half-day workshop (4 hours total, including a ½-hour refreshments break)
Organizers
Dr. Sohan Gyawali
Assistant Professor, Department of Technology Systems
East Carolina University
East 5th Street Greenville, NC 27858 USA
📧 gyawalis22@ecu.edu
đź“ž 252-328-9692
đź”— Website
Dr. Jiaqi Huang
Department of Computer Science and Cybersecurity
University of Central Missouri
116 W South St, Warrensburg, MO 64093
📧 jhuang@ucmo.edu
đź“ž 660-543-8865
đź”— Website
Abstract
Vehicular communication networks enable intelligent transportation systems (ITS) by exchanging Basic Safety Messages (BSMs) to improve safety and traffic flow. Yet they remain vulnerable to attacks—such as denial-of-service, Sybil, false alerts, and insider misbehavior—that traditional cryptographic defenses cannot fully mitigate.
This tutorial presents a unified framework for machine learning–based misbehavior detection (MDS) in next-generation vehicular environments, emphasizing realistic data, privacy, and trust.
We begin with simulation-driven MDS trained on combined local perception and BSM datasets, showcasing its superiority over beacon-only methods. We then explore privacy-preserving techniques: homomorphic aggregation of encrypted feedback for reputation updates without revealing identity or location, and semi-supervised federated learning to train models collaboratively without raw data sharing.
Finally, we address transparency by integrating Explainable AI (XAI) with Large Language Models (LLMs). SHAP values quantify feature contributions, and an LLM uses these insights—plus vector-database contexts—to generate clear, natural-language explanations. This empowers non-expert stakeholders (e.g., network operators, regulators) to understand and trust detection outcomes.
Attendees will leave equipped to design, implement, and evaluate secure, privacy-aware, and explainable MDS solutions for next-generation vehicular networks.
Keywords
- Connected Vehicles
- Misbehavior Detection
- Machine Learning
- Privacy-Preserving Techniques
- Semi-Supervised Learning
- Federated Learning
- SHAP
- LLMs
Tentative Session Program
| Time | Session Description |
|---|---|
| 09:00–10:30 | Session 1: Foundations of MDS and Realistic Data Modeling |
| 10:30–11:00 | Break (Refreshments) |
| 11:00–12:00 | Session 2: Privacy-Preserving MDS: FL + Encryption |
| 12:00–13:00 | Session 3: Explainability with SHAP + LLMs |