Evaluation of VANET datasets in context of an intrusion detection system

Abstract

Vehicular Ad hoc Networks (VANETs) are the core of Intelligent Transportation Systems (ITS), allowing vehicles to communicate between themselves and with other entities. However, these are very complex networks with volatile architectures, ever-changing members, and multiple types of entities, making them an appealing target for attackers since they can find vulnerabilities and perform attacks with massive impact. The most common security approach is to use traditional tools that can help to prevent attacks. Nonetheless, sometimes attacks cannot be prevented, requiring the use of other tools. Intrusion Detection Systems (IDSs) can detect attacks and trigger a response to minimize their effects. IDSs can take advantage of Machine Learning (ML) algorithms to improve their performance, which are excellent at detecting deviations from patterns. However, these need large sets of data to be trained efficiently, but VANET datasets are a scarce resource. This work aims to evaluate the found publicly available datasets in the context of an IDS for VANETs in three stages: 1) First, evaluate the publicly available VANET datasets and gauge their usability in future works; 2) Then, as the datasets found available to the public are divided depending on the geographic region where they were obtained, assess the advantage of using multiple datasets to train an ML algorithm; 3) Finally, the IDS can be located in different network locations (vehicle, a cluster of vehicles, Road Side Unit (RSU), etc.), which directly influences their performance. The impact of the IDS location in the network is also to be studied. The ML algorithm used is the same across all the experiments, maintaining the same basis for all the tests, with the only variable being the dataset.