Date of Award

10-28-2024

Publication Type

Thesis

Degree Name

M.Sc.

Department

Computer Science

Keywords

BERT;Controller Area Network;Intrusion Detection;Machine Learning;Vehicle Security

Supervisor

Ikjot Saini

Creative Commons License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.

Abstract

Electronic Control Units (ECUs) have become important to modern vehicles, enhancing operational control, driving comfort, and safety. These ECUs communicate using the Controller Area Network (CAN) protocol, which despite its widespread adoption, is vulnerable to various security threats. Therefore making CAN protocol more secure is crucial, intrusion detection systems (IDS) offer a viable path to mitigating cyberattacks on vehicles. This research proposes a Hybrid Framework for performance analysis of Machine Learning-based IDS employing a range of models, including K-Nearest Neighbor (KNN), Logistic Regression, Decision Tree (DT), Random Forest, XGBoost, CatBoost, Multi-Layer Perceptron (MLP), Long Short-Term Memory (LSTM), and Bidirectional Encoder Representations from Transformers (BERT). Our research used the newly published can-train-test dataset specifically designed for building ML-based IDS. The dataset consists of nine different types of attacks performed on four different vehicles in a real-world environment. This study majorly focused on DoS, Fuzzy, Gear Spoofing, Speed Spoofing and Standstill attacks performed on Chevrolet Impala and Subaru Forester. Notably, proposed framework, which leverages both traditional Machine Learning models and advanced Deep Learning architectures, achieves good accuracy with less training and prediction time, exhibiting a high true positive rate and a low false negative rate. Additionally, our novel approach involves converting CAN bus data into feature strings for classification using BERT, providing a new way of treating CAN numerical data as string. The models used in proposed Hybrid Framework is evaluated using various performance metrics, including Accuracy, Precision, Recall, F1-score, and AUC-ROC. The results demonstrate that the proposed Hybrid Framework offers a more effective and efficient solution for testing multiple ML approaches in detecting cyberattacks on vehicles

Available for download on Saturday, October 25, 2025

Share

COinS