TransICNN-IDS: Improved convolutional Neural Network with Transformer for intrusion detection system

As network threats become more sophisticated and prevalent, the digital twin technology becomes more and more popular for intrusion detection system (IDS). So it is imperative to design a high accuracy network intrusion detection algorithm suitable for digital twin system. This paper proposes an effective network intrusion detection algorithm TransICNNIDS(Improved convolutional Neural Network with Transformer for intrusion detection), which combines CNN(convolutional neural network) and Transformer structures. Inception block and ECA(efficient channel attention) block are also used to enhance the classification ability of the model. It achieves $\mathbf{9 8. 5 9 \%}$ accuracy, $98.47 \%$ precision, $99.82 \%$ recall, $99.14 \%$ F1 score on the CICIDS2017 dataset and $\mathbf{8 2. 3 6 \%}$ accuracy, $\mathbf{8 1. 6 5 \%}$ precision, 83.72% recall, 84.59% F1 score on the UNSW-NB15 dataset, which is superior to other similar algorithms.