University Events

Description: Topic: Misbehavior Detection in Vehicular Networks With Machine Learning

Zoom Teleconference: https://umassd.zoom.us/j/806213361

ABSTRACT:
Autonomous Vehicle Technology (AVT) offers fundamental restructure of transportation. Connected Vehicles Technology (CVT) further enhances Intelligent Transportation Systems (ITS). Despite the maturity of the technology, the deployment in the real-world standstills partially due to security concerns. IEEE 1609.2 provides security mechanisms by defining secure message formats and procedures. Traditional authentication approaches such as public key infrastructure (PKI) are insufficient due to the rapid dynamics and privacy requirements of Vehicular Ad-Hoc Networks (VANET). By injecting malware to VANET devices or by stealing certificates an attacker can fabricate application or management messages, compromising the security provided with IEEE 1609.2. Existing solutions to secure backend in vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communications include Security Credential Management System (SCMS). It supports certificate provision with privacy preservation and operation efficiency as PKI for Vehicle Networks or "Vehicular PKI".

However, timely misbehavior detection remains an open problem. This dissertation research focuses on enhancing the security of the legitimate vehicles' credential management in Vehicle Networks by ensuring the validity of sent and received data. By utilizing context-adaptive machine learning techniques to secure V2V communication, we present solutions that can timely detect misbehavior caused by insiders performing malicious activities or invalid behavior. The proposed novel solution works in two phases: Phase I: Train the Machine Learning framework integrated with four novel plausibility checks on time-series adversarial dataset generated by an attack engine as well as real time dataset from USDOT Wyoming Connected Vehicle Pilot Deployment and the University of Michigan Transportation Research Institute. Phase II: Test the model and validate it before the next round of model deployment. The results are evaluated using Precision Recall and Receiver Operating Characteristics derived from the Confusion Matrix to validate the effectiveness of the framework. Our overall contribution is three fold: (i) Misbehavior Detection Framework, built by integrating novel plausibility checks along with Supervised Machine Learning techniques. (ii) Real Time Detection with a novel Data Centric misbehavior detection system, built specifically for highway traffic, using Unsupervised Machine Learning techniques. (iii) Introduced Adversarial Attacks using Deep Learning techniques to examine the vulnerabilities in the existing solutions; therefore demonstrating the need of studying attacker strategies to protect the critical infrastructure.

This work focuses on safeguarding Vehicular Networks' security with privacy by studying the topic of misbehavior and its detection. Our results demonstrate that by applying machine learning, data centric analysis accelerates misbehavior detection in a scalable way to ensure safety and security of driverless cars.

Note: All ECE Graduate Students are ENCOURAGED to join the zoom teleconference.
All interested parties are invited to join.

Advisor: Dr. Hong Liu
Committee Members: Dr. Paul J. Fortier and Dr. Honggang Wang, Department of Electrical & Computer Engineering, UMass Dartmouth; Dr. Kavitha Chandra, Department of Electrical & Computer Engineering, UMass Lowell

*For further information, please contact Dr. Hong Liu at 508.999.8514, or via email at hliu@umassd.edu.