A context-aware on-board intrusion detection system for smart vehicles

Abstract

AbstractModern vehicles are becoming more appealing to potential intruders due to two primary reasons. Firstly, they are now equipped with various connectivity features like WiFi, Bluetooth, and cellular connections, e.g., LTE and 5G, which expose them to external networks. Secondly, the growing complexity of on-board software increases the potential attack surface. In this article, we introduce CAHOOTv2, a context-sensitive intrusion detection system (IDS), aiming at enhancing the vehicle’s security and protect against potential intrusions. CAHOOTv2 leverages the vehicle’s sensors data, such as the amount of steering, the acceleration and brake inputs, to analyze driver habits and collect environmental information. To demonstrate the validity of the algorithm, we collected driving data from both an artificial intelligence (AI) and 39 humans. We include the AI driver to demonstrate that CAHOOTv2 is able to detect intrusions when the driver is both a human or an AI. The dataset is obtained using a modified version of the MetaDrive simulator, taking into account the presence of an intruder capable of performing the following types of intrusions: denial of service, replay, spoofing, additive and selective attacks. The sensors present in the vehicle are a numerical representation of the environment. The amount of steering, the acceleration and brake inputs given by the driver are based on the environmental situation. The intruder’s input often contradicts the driver’s wishes. CAHOOTv2 uses vehicle sensors to detect this contradiction. We perform several experiments that show the benefits of hyperparameter optimization. Indeed, we use a hyperparameter tuning paradigm to increase detection accuracy combining randomized and exhaustive search of hyperparameters. As a concluding remark, the results of CAHOOTv2 show great promise in detecting intrusions effectively.