False Information Attack Detection in a Connected Vehicle Environment With Quantum Inspired Long Short-Term Memory

Wireless communication Systems enabling Vehicle-to-Vehicle (V2V) and Vehicle-to-Infrastructure (V2I) data exchange, supported by technologies such as Cellular-V2X (C-V2X), has introduced significant cybersecurity challenges, particularly the threat of false information attacks that can compromise traffic safety and efficiency. In this thesis, the author focuses on identifying false information cyber-attack on V2I, in which vehicles are sending Basic Safety Messages (BSMs) to a Roadside Unit (RSU) and RSUs are collecting, processing and communicating data back to Connected Vehicles (CVs) to support different CV applications. Despite advances in anomaly detection using Long Short-Term Memory (LSTM) networks, these models often struggle with computational efficiency and performance under real-world constraints. Addressing this issue, this study investigates the potential of Quantum Long Short-Term Memory (QLSTM) as a more efficient and effective alternative for detecting false information attacks in CV environments. The objective of the study is to evaluate and compare the performance of classical LSTM and QLSTM models in terms of accuracy, classification confidence, and computational efficiency. Unlike previous studies that primarily rely on extensive data training, the author focusses on the models' robustness and sensitivity across varying classification thresholds to assess their adaptability in real-time cyberattack detection. Key findings of the study indicate that QLSTM consistently outperforms LSTM across different classification thresholds. At the classification threshold of 0.3, QLSTM achieved 96.74% accuracy, 98.67% recall, 95.02% precision and an AUROC of 97.84%, respectively, compared to LSTM’s 93.00% accuracy, 98.36% recall, 90.22% precision and 93.16% AUROC, respectively. At the classification threshold of 0.5, QLSTM further improved its accuracy to 97.8 4%, 98.67% recall, 96.45% precision, and 97.84% AUROC, outperforming LSTM, which reached 95.40%, 98.36% recall, 93.19% precision, and 95.16% AUROC, respectively. Additionally, QLSTM demonstrated a 99% reduction in trainable parameter count and memory usage, requiring only 195 parameters (0.72 KB) compared to LSTM’s 19,777 parameters (77.21 KB). These results highlight QLSTM’s ability to achieve a high detection accuracy while significantly reducing computational cost. The findings from the study have important implications for the future of CV cybersecurity, demonstrating that quantum-inspired models offer a promising avenue for enhancing real-time anomaly detection. By leveraging quantum principles, QLSTM can provide a more scalable and efficient solution for securing next-generation vehicular network against cyber threats.