IFLV: Wireless network intrusion detection model integrating FCN, LSTM, and ViT

Abstract

Wireless networks are vulnerable to various network attacks due to easy access to the nodes. The development of technologies for network intrusion detection, including those based on deep learning, is expected to bring ultimate solutions to this problem. Nevertheless, existing intrusion detection models based on deep learning have low detection accuracy and cannot effectively detect several new types of attacks. Aimed at such, this article proposes IFLV, an intrusion detection model for wireless networks, by integrating Fully Convolutional Network (FCN), Long Short-Term Memory (LSTM), and Vision Transformer (ViT). IFLV can extract the local and global features of traffic data and learn its temporal and spatial features, to improve the accuracy of network traffic classification. Based on the improvements of the traditional ViT model to overcome the poor classification effect in small and medium-sized datasets, IFLV can achieve expressive results even with fewer training resources. Experimental results show that IFLV has a high accuracy of network traffic intrusion detection with an accuracy of 99.973% in the AWID dataset and significantly superior performance compared to existing models.