Network Intrusion Detection Based on Feature Selection and Transformer

Network intrusion detection plays an important role in the internet with huge traffic. With the rise of deep learning, scholars have begun to conduct intrusion detection based on the powerful feature extraction capabilities of deep learning and its robustness, such as convolutional neural network or Long Short-Term Memory, however, this kind of method is weak in the perception and representation of complex feature samples, resulting in poor final effect. An intrusion detection method based on feature selection and Transformer model is proposed in this paper. Firstly, the features that are most advantageous to model detection are selected by univariate feature selection and feature recursive elimination Secondly, based on Transformer architecture, an intrusion detection model is established, which encodes the traffic characteristics through the encoder in Transformer, and then classifies them using neural network, a fusion loss function is designed to accelerate the convergence of the model and improve the performance of the model. These innovative methods and the optimization of the network structure make the model achieve excellent performance, combining the performance of the model on the public data set and the self-built data set, it surpasses the current most advanced method and is an excellent intrusion detection method.