Securing Internet of Things: An Enhanced Vision Transformer for Intrusion Detection

Intrusion Detection Systems (IDS) are essential for protecting Internet of Things (IoT) networks, which face challenges such as high data volume, evolving attack patterns and limited computing resources. Traditional IDS approaches tend to make mistakes in determining rare or unusual types of attacks, which shows the need of a more accurate and efficient solution. To counter these challenges, this paper proposes a Vision Transformer-based framework, named Enhanced-ViT, which transforms tabular data to RGB images for better spatial feature extraction capabilities of Vision Transformer (ViT) model. This will hence improve the intrusion detection. The proposed method used the CIC IoT 2023 dataset, which for this research, contains nine categories of Denial-of-Service (DoS) and Distributed-Denial-of-Service (DDoS) attacks. Enhanced-ViT achieved an accuracy of 97.65% on the training set and 97.39% on the testing set. The precision and F1-scores were consistently high across most classes. The model's robustness and efficiency can be seen by its ability to detect complex and low-frequency attack patterns, while having a lightweight model design. Main contribution of this work is using the ViT model in a different and innovative way on image-transformed tabular network traffic data for intrusion detection, thus offering an accurate solution with high scalability and reliability when compared to other baseline models.