Position Falsification Attack Detection In Inter-Vehicle Networks Using Deep Learning

Vehicular Ad-hoc Networks (VANETs) is a core component of the Intelligent Transportation System (ITS). It promises to enhance transportation by facilitating coordinated actions among vehicles, infrastructure, pedestrians, and other mobile entities on ad-hoc basis. VANET is expected to disseminate data securely over information-relevant Regions of Interest (ROIs), and guarantee the integrity of data even with the presence of malicious vehicles. Due to VANET’s ad-hoc nature, it is exposed to various cyberattacks, including position falsification. Thus, Intrusion Detection Systems (IDS) play major role in improving the security of VANET. In this paper, a Deep Learning-based IDS (DL-IDS) is constructed to detect various types of position falsification attacks in a decentralized manner. The system employs Multi-Layer Perceptron (MLP) algorithm that considers the RSSI aggregation of first-hop neighbors and the Time Difference of Arrival (TDoA) as new false position detection features, in addition to commonly used parameters. VeReMi data set is used to train the DL-IDS model offline, which can then be deployed at an Onboard Unit (OBU) of a vehicle, thereby reducing the system’s computational complexity and execution time. The DL-IDS model shows high accuracy and F1-score values in detecting three different position falsification attacks and a false emergency breaking alert. The accuracy of the DL-IDS is shown to exceed those of existing models by (2-7)% with a clear advantage in computational complexity.