Advances, Systems and Applications
Table 1 A very short summary of the key points in the related works
From: Security strategy for autonomous vehicle cyber-physical systems using transfer learning
Ref. | Key Attributes | Strengths | Weaknesses |
|---|---|---|---|
[23] | - Hybrid IDS using MLPs and POS techniques - Detection of DoS attacks in autonomous vehicles | - High detection rates for DoS attacks - Experimental validation | - High inferencing overhead - Computational processing through diverse subsystems |
[26] | - Detection method for false data injection - Use of LSTM deep networks | - Superior detection rates - Categorization of data samples as normal or abnormal | - Use of simulated dataset without real-world implementation |
[29] | - ML-IDS to minimize traffic disturbance and collisions - Identification of spoofing and jamming attacks | - Four learning techniques used - DF technique achieved over 90% accuracy | - Detection accuracy needs improvement compared to other models |
- ML-IDS for detecting malicious traffic in the CAN bus network - Naïve Bayes and random decision trees used | - Scalable and adaptable to various attacks | - Utilization of imbalanced classes in the dataset | |
[33] | - ML-IDS for autonomous vehicle communication - Based on ToN-IoT dataset - Use of the Chi-square method for feature selection | - Performance of eight supervised ML techniques evaluated - XGBoost outperformed other methods | The best model may need more sparse and unstructured data |
[37] | - Cloud-based intrusion detection framework - Clustering of autonomous vehicles - Use of DPN and decision tree techniques | - Continuous service availability - High positive and low false detection rates | - Specific to cloud-based detection |
[38] | - Cost-effective, lightweight intrusion detection scheme - Three-layer neural network with Softmax classifier | - Detection accuracy, recall, precision, and F-1 score reported - Outperformed other classification schemes | - Shallow model may struggle with large-scale attack vectors |
[41] | - Hybrid anomaly-based IDS for electronic control units - Rule-based and supervised learning-based methods | - High identification ratio - Minimal computational complexity | - Specific to anomalous CAN communications |
[42] | - Hybrid anomaly-based IDS with rule-based and machine-learning IDS - Decision tree, random forest, and XGBoost used | - Detection of malicious CAN traffic with over 90% accuracy | - Unable to detect attacks leveraging the periodicity of CAN messages |
[44] | - PChain IDS based on Blockchain Technology - Focus on privacy and trust issues in autonomous transportation systems | - Evaluation revealed high detection accuracy - Use of blockchain for security | - Evaluation based on an old dataset of common cyber-attacks |
[45] | - Cooperative IDS for autonomous vehicular networks - Decentralized federated-based methodology - Use of blockchain for storing and sharing training loads | - Decreased resource utilization overhead - Distributed edge devices for training load | - Vulnerability of blockchain to cyber-attacks like ransomware |
[61] | - Exploration of user check-in data for POI recommendations - Critique of recurrent and graph neural network-based methods | - Introduction of ITGCN for successive POI suggestions - Experimental results show improved accuracy | - Limitations of existing methodologies highlighted |
[62] | - Addresses cybersecurity risks in IoV due to CAVs - Proposes IIDS with modified CNN for AV classification - Operates in a 5G V2X environment | - Efficiently identifies and classifies malicious AVs - Enhances traffic safety monitoring and collision avoidance - Achieves 98% accuracy in attack detection | -Limited Security Measures - Focusing on specific CNN hyperparameters may limit its adaptability to emerging cyber threats |
[63] | - Focuses on cyberattacks on in-car networks due to electronics integration - Targets vulnerabilities in the CAN bus - Proposes DCNN-based IDS for the CAN bus | - DCNN learns network traffic patterns without human feature design - Outperforms traditional machine-learning techniques - Demonstrates lower false negative and error rates | - Limited information on the scale or diversity of experiments |
[64] | - Emphasizes IDS necessity for the vulnerable CAN bus.—Introduces GIDS, a GAN-based IDS for in-vehicle networks - GIDS learns and identifies unknown attacks with regular data | - GIDS achieves excellent detection accuracy for unknown assaults - Addresses the lack of security measures in the CAN bus | - Needs more details on the specific architecture of GIDS |
[65] | - Discusses limitations of RNNs in in-vehicle intrusion detection - Proposes TCAN-IDS, a Temporal Convolutional Network with Global Attention - Addresses real-time monitoring challenges | - TCAN-IDS encodes 19-bit characteristics for real-time monitoring - Global attention mechanism enhances feature extraction - Demonstrates strong detection performance on known attack datasets | - No details were provided on the scale or diversity of experiments |
[66] | - Highlights difficulties in intrusion detection for vehicle communications - Proposes STC-IDS for spatial–temporal correlation - Addresses challenges in local feature consideration and poor multi-feature mapping | - STC-IDS encodes spatial and temporal interactions simultaneously - Outperforms baseline approaches - Achieves lower false alarm rates while maintaining efficiency | - Needs more specific details on the architecture of STC-IDS |