Securing the Road: Advancing Cybersecurity in Internet of Vehicles with Deep Learning

Abstract

The Internet of Automobiles (IoA) facilitates the exchange of safety-related messages among vehicles, thereby reducing road accidents. Nevertheless, this communication network is susceptible to a number of threats, including erroneous alerts and mispositioning of the vehicle. This paper addresses challenge of authenticating messages to distinguish normal packets from attack packets by proposing an approach to deep learning with binary classification. We utilize Rectified Linear Unit (ReLU) activation algorithms in conjunction with SoftMax classifiers for structured deep neural network to classify normal and malicious packets. The training dataset, prepared from KDD99 and CICIDS 2018 datasets, comprises 120,000 network packets with more than 40 features. Initial preprocessing involves using an autoencoder to eliminate irrelevant data, resulting in 22 valuable features. The Deep Neural Network (DNN) model is trained using Google Colab, utilizing TensorFlow, and validated using a simulated dataset produced via network simulation. Accuracy of investigational findings is 99.48%, that is higher than current models built using convolutional and recurrent neural networks (RNN) and (CNN), respectively. Incorporating sophisticated anomaly detection methods with reinforcement learning approaches may present interesting directions for future study, improving the flexibility and resilience of car communication safety features in ever-changing Internet of Things.The research's primary findings highlight the deep learning-based techniques  potential to greatly improve the security and dependability of vehicular communication networks, opening the door for more secure and robust transportation networks in the age of the Internet of Automobiles.