LSTM-Based Hybrid Intrusion Detection System for Internet of Vehicles

The recent growth of the Internet of Things (IoT) has revolutionized vehicular networks into the Internet of Ve-hicles (IoV). Within the IoV infrastructure, modern vehicles are vulnerable to different and new types of cyber-attacks. Consequently, Intrusion Detection Systems (IDS) are extremely helpful in coping with such attacks, and there is a pressing need for the development of an advanced IDS that can efficiently detect attacks with a high detection rate and accuracy, and a low false alarm rate. Toward this end, we present a deep learning-based IDS for the identification and classification of potential cyber-attacks in the 10 V network. Specifically, a generative hybrid deep learning model that combines Long Short-Term Memory Variational AutoEncoder with Bidirectional Gated Recurrent Units (LSTMVAE-BiGRU) is proposed. Due to the integration of the LSTM network and the VAE in this architecture, the LSTM performs the encoding and decoding in the LSTMVAE paradigm. The LSTMVAE is forced to learn the most salient features of the training time series data by restricting the latent space to have a dimension that is less than the input. On the other hand, to learn latent representations, BiGRU collects local dependencies in a two-way time flow, or forward and backward. Lastly, the multi-class attack detection method is carried out using the softmax layer. Experimental findings demonstrate that the proposed IDS outperforms both baseline techniques that are widely used and cutting-edge approaches.