An Attention-Based CNN with Batch Normalization Model for Network Intrusion Detection

Abstract

The dynamicity of network environment and the diversity of network attacks cause huge challenges for network security. Network intrusion detection technology plays an important role in improving network security. In this paper, based on Convolutional Neural Networks, we propose a novel network intrusion detection model (ACNNBN). The ACNNBN introduces Convolution Block Attention Module to realize cross-channel fusion with convolutional module, which allows the ACNNBN to quickly select information that is more critical to the current task. In addition, the model can learn some significance feature of the shallow layer that the deeper layer of CNN may lose, and prevent vanishing gradient and accelerate model convergence. It greatly improves the processing efficiency and accuracy of network traffic features. We use CSE-CIC-IDS2018 dataset to evaluate the effectiveness of the model. For comparison study in performance, we compare our model to other five algorithms, and select evaluation indicators to test the performance of different models. Experimental results show that the ACNNBN model can effectively enhance the detection accuracy, and which performance is distinctly superior to the other five models in the evaluation index.