An innovative feature selection approach for CAN bus data leveraging constant value analysis

Althunayyan, Muzun, Javed, Amir ORCID: https://orcid.org/0000-0001-9761-0945 and Rana, Omer ORCID: https://orcid.org/0000-0003-3597-2646 2024. An innovative feature selection approach for CAN bus data leveraging constant value analysis. Presented at: AI Applications in Cyber Security and Communication Networks, Cardiff, UK, 1–12 December 2023. Published in: Hewage, C., Nawaf, L. and Kesswari, N. eds. Proceedings of Ninth International Conference on Cyber Security, Privacy in Communication Networks. Lecture Notes in Networks and Systems , vol.1032 Singapore: Springer, p. 69. 10.1007/978-981-97-3973-8_5

Preview

PDF - Accepted Post-Print Version Download (953kB) | Preview

Abstract

Intrusion detection systems (IDS) serve as effective security measures within in-vehicle networks. There is a growing demand for lightweight and computationally efficient IDS solutions compatible with systems constrained by limited computing and storage resources. One way to achieve this is to use feature selection methods to reduce computational costs. However, selecting a subset of features introduces a potential vulnerability, allowing adversaries to exploit unselected features for new unknown attacks and bypass IDS. To address this concern, we propose a novel observation-driven feature selection approach for CAN bus data. This approach selects the most important features without losing valuable information and prevents adversaries from exploiting unselected features. We validate our observations using three benchmark datasets. We assess the impact of our proposed approach on the number of trained parameters, false positives, false negatives, and F1-score. We illustrate how our approach effectively addresses the risks associated with adversaries exploiting unselected features. Experimental results demonstrate that our approach reduces the number of trained parameters by approximately 44% in a machine learning model and by 14.24% in a deep learning model. Moreover, the results show that our approach helps the model detect around 8% of unknown attacks. Our approach reduces computational overhead, thereby improving overall computational efficiency. It demonstrates promising results by reducing computational resources and minimising their vulnerability to potential malicious traffic injection, thereby enhancing vehicle security.

Item Type:	Conference or Workshop Item (Paper)
Date Type:	Publication
Status:	Published
Schools:	Computer Science & Informatics
Publisher:	Springer
ISBN:	9789819739721
ISSN:	2367-3370
Date of First Compliant Deposit:	19 September 2024
Date of Acceptance:	21 December 2023
Last Modified:	02 Oct 2024 14:15
URI:	https://orca.cardiff.ac.uk/id/eprint/172253

Actions (repository staff only)

Edit Item