Hybrid quantum machine learning architecture for detecting DoS and fuzzy attacks in intelligent vehicle CAN networks

Abstract

This study introduces a hybrid quantum machine learning (QML) model integrating a quantum convolutional autoencoder (QCAE) and a quantum orthogonal classifier with support vector machine (QOC–SVM) for detecting denial-of-service (DoS) and fuzzy attacks in the controller area network (CAN) of autonomous vehicles (AVs). The framework employs quantum feature encoding and orthogonal transformation to extract complex patterns from vehicular communication data. Using both public and simulated datasets generated via the CARLA 0.9.14 simulator, the model achieved an accuracy of 99.38%, precision of 0.990, recall of 0.988, and an F1-score of 0.993 at a batch-to-batch size ratio of 7700:30. Compared to traditional models like random forest (RF) (89%) and convolutional neural networks (CNN) (94.5%), the proposed hybrid system exhibited superior detection accuracy. The learning curve showed smooth convergence without overfitting, validating its stability and scalability. This approach highlights the promise of quantum-classical intrusion detection systems (IDS) for real-time cybersecurity in AVs.