Date of Award

2-4-2025

Publication Type

Thesis

Degree Name

M.Sc.

Department

Computer Science

Supervisor

Ikjot Saini

Rights

info:eu-repo/semantics/openAccess

Abstract

Vehicular Ad hoc Networks (VANETs) enable communication between vehicles and roadside infrastructure to improve traffic efficiency and road safety. However, their susceptibility to security threats, particularly Denial of Service (DoS) attacks, poses significant challenges to network entities, preventing access to critical resources and services.One of the major reason that DoS attack poses great threat to VANETs is because they can be launched in different forms . In this thesis, we propose a lightweight hybrid approach that combines deep learning and machine learning models. This approach automates feature engineering tasks such as feature selection, noise reduction, and the discovery of hidden temporal and spatial dependencies, while also preserving data privacy with the use of deep learning models. By eliminating unnecessary communication data, the model enhances the efficiency of attack classification while reducing computational complexity. Our framework addresses multiclass classification, effectively identifying five distinct types of DoS attacks within a unified system. Unlike traditional methods focusing on single attack types, our proposed models are designed for real-world applicability, incorporating novel evaluation metrics such as time and space complexity. The best-performing model achieves a time taken of 8.9 × 10−7 seconds and a space complexity of 44.514 MB, making it lightweight and deployable on On-Board Units (OBUs) in vehicles.

Download

Share

COinS