Explainable Deep Learning Models for Intrusion Detection in Autonomous Vehicle Networks

Abstract

The anomaly classification component applies decision-making tree (DT) classifier and generates interpretable rules related to adversarial instances. The subsequent verification (VM) component uses sequential feature from data instances that are classified as adversarial to check whether a packet is benign or false. The HEDL model allows the use of suitable algorithms wherein the ID part can use hidden Markov Model (HMM), conditional restricted boltzmann machine (CRBM) and long short-term memory (LSTM) algorithms. The open-source NSL–KDD dataset and NASA datasets are leveraged for training and testing of the proposed model. The experimental results demonstrate that the proposed hybrid model offers improved results including precision, recall and accuracy as compared with various existing hybrid and sequential models in the literature [1] . The model also allows for interpretation by means of easy to understand rules and unidentified attack services can be detected through ID agent which is not available in majority of the existing models. Future scientific work covers verify the performance of the proposed model on real data by means of software-in-the-loop (SiL) in vehicle network.