Deep Learning Model based IDS for Detecting Cyber Attacks in IoT based Smart Vehicle Network

The Internet of Vehicle (IoV) is a technology that connects vehicles such as cars and buses to the internet. It facilitates the exchange of information between these vehicles for increased safety, efficiency, and speed. The IoV architecture consists of three layers: perception, network, and application. The main challenge in the design of IoT is the security issue. To handle the security issue we proposed an Intrusion Detection System (IDS) that uses deep learning models, specifically deep neural networks, to identify malicious network traffic and to classify the type of attack. The system operates in two stages: in the first stage, normal and attack traffic are differentiated, and in the second stage, the type of attack is determined. Three variants of rule extraction methods are proposed here including homogeneous methods using either decompositional or pedagogical rule extraction in both stages, and a heterogeneous method using pedagogical for binary classification and decompositional for attack classification. The goal is to achieve improved classification accuracy while reducing resource consumption. The proposed IDS was tested using car hacking datasets for both external and in-vehicle communications. The results indicate that the deep learning based decompositional with pedagogical method achieved the best results in all cases of the IDS system with high accuracy under the Car-hacking dataset.