Incremental Development of a Multilayer Neural Network for CAN bus Attack Detection

The growing integration of electronic systems in vehicles has transformed the automotive industry, enabling improved performance, reduced costs, and enhanced functionality. However, the increased complexity of vehicular networks and the lack of built-in cybersecurity mechanisms in communication protocols such as the Controller Area Network (CAN) bus have left modern vehicles vulnerable to cyberattacks. This paper presents a comprehensive investigation on intrusion detection in the CAN bus using multilayer neural networks (MLPs). A process is presented to find a neural network capable of detecting three types of attacks. This consisted of a series of experiments to find the minimal structure of a neural network that can detect malicious traffic on the CAN bus, allowing the identification of attack-free data and data with three types of attacks, denial-of-service (DoS), impersonation, and fuzzy attacks. Additionally, this work contextualizes CAN security issues, discusses related contributions, and presents future research opportunities. Smart citations: https://scite.ai/reports/10.61467/2007.1558.2026.v17i2.1263Dimensions.Open Alex.