A comparative analysis of anomaly detection methods in V2X communication network under smart cities

Smart cities are characterised by a combination of advanced technologies and innovative strategies aimed at enhancing residents' quality of life. In this paper, we compare 6 anomaly detection methods on two commonly used streaming data sets. The tool called HoneyThing employs the Honey Badger Algorithm for anomaly detection. Data points that significantly differ from the rest are considered anomalies. The paper emphasises the importance of anomaly detection techniques in data science and discusses the limitations of centralised machine learning approaches and cross-chain anomalies in a multi-blockchain IoT ecosystem, including the V2X Communication Networks. Existing anomaly detection systems need improved accuracy due to inefficient feature extraction and selection procedures. Traditional systems in V2X (Vehicle-to-Everything) networks often rely on signature-based detection methods that are ineffective against zero-day attacks or previously unknown anomalies. This study provides a comparative analysis of various methods for detecting misbehavior or anomalies in smart cities, including Statistical Mahalanobis Distance, Integrated Honeypot, Honeything, Honeywall with CDQL Algorithms, HoneyThing with Autoencoders and XGBoost, ML with Python Fast API, Federated Detection Methods and GAN (Generative Adversarial Networks). The research encourages further efforts to address some of the challenges of anomaly detection in 6G V2X communication networks within smart cities by identifying the most effective among the five discussed methods. Results show that the approach combining Autoencoder with the Honey Badger Algorithm (HBA) significantly outperforms traditional machine learning techniques such as XGBoost and ensemble methods. The (Autoencoder + HoneyThing + XGBoost) approach achieves the highest AUC (~ 0.85), indicating its superior ability to detect anomalies with minimal false positives. The models were evaluated on four different datasets using four evaluation metrics to assess performance from various perspectives, like—our analysis confirms that integrating an Autoencoder with Honeything and XGBoost outperforms traditional anomaly detection methods. Cross-validation techniques are additionally recommended to offer a more precise assessment of the model’s performance. In this study, there are various methods mentioned in this paper wherein V2X and C-V2X Communication are suggested under smart cities, and the outliers or anomalies are detected based on specific design and evaluation decisions. Below are some of the highlights of this article: