Journal
IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS
Volume 34, Issue 9, Pages 2512-2528Publisher
IEEE COMPUTER SOC
DOI: 10.1109/TPDS.2023.3290650
Keywords
Training; Intrusion detection; Security; Vehicular ad hoc networks; Collaboration; Data models; Computational modeling; Federated deep learning; collaborative intrusion detection system; intelligent transportation system; Index Terms; convolutional neural network; gradient optimization
Ask authors/readers for more resources
With the development of communication technology and intelligent transportation systems, vehicular ad hoc networks (VANETs) have gained popularity, leading to increased importance of VANET communication security. An intrusion detection system (IDS) is essential in improving network security, but faces challenges in accuracy, efficiency, and completeness due to frequent location changes in VANETs. This study proposes a collaborative intrusion detection system (CIDS) model that utilizes federated learning in software-defined VANETs to address these issues. The model trains through collaboration among local software-defined networks (SDNs) without sharing local network data flows, improving IDS scalability and globality.
With the continuous innovations and development in communication technology and intelligent transportation systems, a new generation of vehicular ad hoc networks (VANETs) has become increasingly popular, making VANET communication security increasingly important. An intrusion detection system (IDS) is an important tool for detecting network attacks and is an effective means of improving network security. However, existing IDSs encounter several problems involving inaccurate detections, low detection efficiencies, and incomplete detections owing to extensive changes in vehicle locations in VANETs. This study explores federated learning in software-defined VANETs and designs an efficient and accurate collaborative intrusion detection system (CIDS) model. The model utilizes the collaboration among local software-defined networks (SDNs) to jointly train the CIDS model without directly exchanging local network data flows to improve the expansibility and globality of IDSs. To reduce the model difference between different SDN clients and improve the detection accuracy, this study regards the prediction loss for each SDN client as an objective from the perspective of constrained multi-objective optimization. By optimizing a surrogate maximum function containing all the objectives, the method adopts two-stage gradient optimization to achieve Pareto optimality for SDN clients with the worst fairness constraint maximization performance. In addition, this study evaluates the training model using two open-source datasets and compares it with the latest methods. Experimental results reveal that the proposed model ensures local data privacy and demonstrates high accuracy and efficiency in detecting attacks and is thus superior to the current schemes.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available