Date of Award

7-8-2024

Publication Type

Thesis

Degree Name

M.Sc.

Department

Computer Science

Keywords

MACHINE LEARNING;MISBEHAVIOUR DETECTION;POSITION FALSIFICATION ATTACK;UNSUPERVISED ALGORITHMS;VANETS;VEREMI DATASET

Supervisor

ARUNITA JAEKEL

Abstract

A wireless network connects a collection of moving or stationary automobiles and other infrastructure nodes to form a vehicular ad hoc network or VANET. One of the primary functions of VANETs is to ensure drivers' comfort and safety in moving vehicles. Vehicles in VANET share safety and non-safety information with each other through periodic broadcasts of Basic Safety Messages (BSM), which contain the vehicle's relevant status information, such as position, heading, speed, etc. BSMs are used for safety applications such as collision avoidance and inserting false information in BSMs can have serious consequences. An attack known as position falsification occurs when a vehicle broadcasts a fictitious BSM position, which can cause traffic jams or even accidents and such attacks need to be detected quickly and accurately. A number of supervised learning algorithms have been proposed for detecting such attacks. However, it is difficult to obtaining labelled datasets in "real-world" scenarios, so there is a need to design suitable unsupervised ML models. In this thesis, we have compared four unsupervised Machine Learning models performance for detecting position falsification. The effectiveness of the proposed models have been evaluated by comparing with existing detection techniques using the publicly available Vehicular Reference Misbehavior (VeReMi) dataset, based on simulated traffic data.

Download

Share

COinS