Blockchain-based federated learning framework for malicious node detection in internet of vehicles (IoV) networks using fog and cloud computing

Download PDF
Open Webpage

Srinivas Reddy Bandarapu, Muhammad Bilal, Pushpalika Chatterjee, Adnan Mustafa Cheema, Junaid Rashid, Jungeun Kim

Published: 2025, Last Modified: 23 Jan 2026J. King Saud Univ. Comput. Inf. Sci. 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Due to the continuous digitalization, IoV networks are vulnerable to various communication attacks by malicious network nodes. In these attacks, the malicious entities disseminate faulty information in the network, which affects quick and intelligent decision-making in the network. Many deep learning and machine learning techniques are proposed for the classification of legitimate and malicious vehicular entities. These techniques have a centralized model training structure, which has low classification accuracy and is vulnerable to privacy leakage. To address these issues, we propose a blockchain-based federated learning framework for distributed classification of malicious and legitimate vehicles. The proposed model uses the capabilities of Long short-term memory (LSTM) and Naive Bayes (NB) for efficient and reliable malicious node detection. In our proposed model, the distributed models are trained on each locally installed virtual machine with a federated learning mechanism and then a unified model is generated at the centralized cloud server. The proposed model not only enhances the accuracy and privacy preservation but also solves the issues of centralized Internet of Vehicles (IoV) networks such as single point of failure and performance bottlenecks by utilizing the capabilities of blockchain. We used the Vehicular Reference Misbehavior (VeReMi) dataset for evaluation of our proposed model. The results show that our proposed LSTM and NB-based model outperforms centralized benchmark classification methods in malicious node detection. With an accuracy of 95%, the LSTM-based model demonstrates superior performance in identifying both malicious and legitimate vehicles, achieving a precision of 0.96 and a recall of 0.97. The high value of precision and recall shows that our model can efficiently discriminate between malicious and legitimate vehicles in the IoV network.