DGA domain name detection based on BiGRU-MCNN

Abstract

Aiming at the problem that DGA domain name is difficult to detect effectively, a hybrid model based on bidirectional gated recurrent unity and multi-channel convolutional neural network is proposed for DGA domain name detection. The model consists of three parts: the character embedding layer, the feature extraction layer and the classification prediction layer. The character embedding layer completes the automatic encoding of the input characters; the feature extraction layer uses BiGRU to learn the dependency between data features and the MCNN uses different neural network channels to learn information from various aspects to obtain deep hidden information for automatic extraction. The characteristics of the input characters; the classification layer uses a three-layer fully connected neural network to achieve automatic prediction classification of DGA domain names. The experimental results show that the model achieves an accuracy of 92.27%, which improves the accuracy of detection compared with other deep learning methods.