DoS Attack Detection in VANET using Transfer Learning Approach for BSM Data

In vehicular ad hoc networks (VANETs), vehicles exchange sensitive information, including vehicle identity, position, speed, heading and other parameters, with nearby vehicles or infrastructure through Basic Safety Messages (BSM). However, VANET communication is vulnerable to various types of attacks, and appropriate security mechanisms must be in place to ensure that exchanged messages are not altered or false messages created by malicious attackers. In this paper, we propose a new transfer learning (TL) based technique for detecting different types of denial-of-service (DoS) attacks in vehicle-to-vehicle (V2V) communication. We start with a pre-trained CNN-BiLSTM model, which was trained to detect various types of attacks in internet traffic. We update the model by training the LSTM layers, while keeping the weights of the CNN layers unchanged. The proposed TL approach is not only able to detect malicious BSMs/senders with higher accuracy but also requires less training time compared to the same DL algorithms with no transfer learning. Simulation results demonstrate that our approach outperforms the existing techniques for each of the different types of DoS attacks in the VeReMi extension dataset in terms of accuracy, precision, recall and F1-score.