Intrusion Detection in Smart Vehicles

Abstract

The Internet ofVehicles (IoV) istransforming transportation by enabling real-timecommunication betweenvehicles andinfrastructure.AcoreelementofthiscommunicationistheControllerAreaNetwork(CAN)bus,whichinterconnectsElectronicControlUnits(ECUs)withinvehicles.However,theCANbushassignificant security vulnerabilities, making itsusceptible to cyberattacks that could disruptvehicleoperations. Currentmethodsforintrusiondetectioninvehicles include rule-based systems, signaturedetection, and anomaly detection. Rule-basedsystemsrelyonpredefinedrulestodetectknownattackpatterns,whilesignaturedetection identifies specific attack signaturesinnetworktraffic.Anomalydetection,incontrast,flagsdeviationsfromtypicalbehavioras potentialintrusions. Thesetraditionalmethodsfacelimitations,includingdifficultiesinidentifyingneworevolving attack patterns and a high rate offalsepositives.Toaddressthesechallenges,theiyproposedtheOptimalAttentionDeepLearning-basedIn-vehicleIntrusionDetectionand Classification algorithm (OADL-IVIDC).Leveragingtheattention mechanism, thissystemfocuses oncriticaldata, enhancingaccuracy indetectingbothknownandunknownthreatswhilereducingFalsepositives. This deeplearning-based approachprovidesamoreadaptiveandreliablesolutionforsecuringCANbuscommunicationin theIoVenvironment.

Introduction

The rapid advancement of Intelligent Transportation Systems (ITS) and Internet of Vehicles (IoV) has increased vehicles’ reliance on Controller Area Network (CAN) buses for communication, which were originally designed without security in mind, exposing them to cyberattacks. Traditional Intrusion Detection Systems (IDS) based on classical machine learning or even some deep learning models struggle to effectively detect and classify diverse CAN attacks in real time due to lack of temporal awareness, interpretability, and robustness against obfuscated attacks.

To overcome these limitations, the text proposes an Optimal Attention-based Deep Learning Intrusion Detection and Classification system (OADL-IVIDC), which integrates an Attention-enhanced Long Short-Term Memory (A-LSTM) model. This system emphasizes critical temporal features in CAN traffic, allowing accurate detection and classification of multiple attack types such as DoS, Fuzzy, Gear, and RPM attacks. The model is supported by thorough data preprocessing, hyperparameter tuning (using RMSProp optimizer), and extensive benchmarking against other classifiers.

Experimental results on a benchmark car hacking dataset demonstrate that OADL-IVIDC achieves high accuracy (99.78%), precision (99.79%), recall (99.80%), and strong performance in distinguishing attack types. Compared to previous works (including SVM, LSTM without attention, GAN-based IDS, and traditional RNNs), the proposed system shows superior detection accuracy, reduced false positives, better interpretability, and real-time applicability.

The architecture also enables real-time alerting and defensive actions to enhance vehicle cybersecurity and passenger safety, representing a significant advancement in IDS for smart vehicular networks.

Conclusion

The OADL-IVIDC system presented in this paper offers an innovative solution to enhancing the security of modern vehicles by detecting and classifying intrusions on the Controller Area Network (CAN) bus. By integrating an Attention-based Long Short-Term Memory (A-LSTM) model with an attention mechanism, the system effectively captures the temporal dependencies in CAN message sequences, which is critical for accurate intrusion detection. The model achieved outstanding performance with 99.78% accuracy, 99.80% detection rate, and high precision and recall across various types of attacks, including DoS, Fuzzy, RPM, and Gear attacks. The results demonstrate that the proposed system significantly outperforms traditional machine learning models and even other deep learning architectures, such as CNN-based models, in detecting intrusions on the CAN bus. This highlights the efficacy of combining attention mechanisms with LSTMs for this type of cybersecurity application.

References

[1] Avapour, O., et al. (2020). \"Intrusion Detection Systems for In-Vehicle Networks: A Survey.\" Journal of Cybersecurity, 12(3), 99-115. [2] Xie, H., & Zhang, J. (2018). \"A Deep Learning Approach for Intrusion Detection in Controller Area Network.\" IEEE Access, 6, 45678-45689. [3] Choi, Y., et al. (2019). \"Cybersecurity in Automotive CAN Bus Networks: Challenges and Solutions.\" Proceedings of the IEEE International Conference on Cyber Security, 45-53. [4] Lee, S., & Kim, D. (2021). \"An Efficient Intrusion Detection System for CAN Bus Using Convolutional Neural Networks.\" IEEE Transactions on Intelligent Transportation Systems, 22(5), 2309-2320. [5] Akhavan, A., et al. (2020). \"Security Challenges in the IoT-Connected Automotive Ecosystem.\" IEEE Internet of Things Journal, 7(12), 11301-11313. [6] Zhang, Y., et al. (2021). \"A Survey on Machine Learning for Intrusion Detection Systems in Cyber-Physical Systems.\" IEEE Access, 9, 12745-12760. [7] Cui, W., & Xu, C. (2020). \"Intrusion Detection for CAN Bus Using Deep Neural Networks.\" Journal of Network and Computer Applications, 169, 102748. [8] Peres, P., et al. (2022). \"Real-Time Intrusion Detection in CAN Bus: A Review and Classification Framework.\" Journal of Cyber-Physical Systems, 8(4), 543-558. [9] Wang, L., et al. (2019). \"Attention Mechanisms in Deep Learning for CAN Bus Anomaly Detection.\" Proceedings of the 2019 IEEE International Conference on Cyber Intelligence and Security, 215-220. [10] Kumar, S., & Singh, A. (2021). \"Evaluation of Machine Learning Techniques for CAN Bus Intrusion Detection.\" International Journal of Computer Applications, 175(3), 33-41.

Copyright

Copyright © 2025 Mrs. D Suganya, Dr. Suresh Kumar R G, Alan Sharon Joseph Fernandez, Aswin Valsaraj, Mohammed Nehal T, Muhammed Rafin Asharaf T. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.