Detection for domain generation algorithm (DGA) domain botnet based on neural network with multi-head self-attention mechanisms

Abstract

Domain Generation Algorithm(DGA) serve as variety detection of domain utilized with malicious into develop random domain as a specific algorithm that accept assignation parts on that malicious botnet communication. In recent days, DGA has presented in addition to communicated extensively harder identifying through these classical feature based DGA. By this chapter, they utilize these character-based to these DGA domain together with present a framework of Deep Neural Network to classifying DGA and existing methods by DGA identification utilized machine learning approach predicated on characteristics sampling in addition to deep learning approach on identify botnet domain. At the same time, these techniques achieve weakly performance against the word-list based botnet. In this chapter, the proposed deep learning approach of Multi-Head Self-Attention-Recurrent Convolution Neural Network-Self Attention Bidirectional Long Short Term Memory model toward identifying together with classifies DGA domain threat. Initially, this Recurrent Convolution Neural Network along with Self Attention Bidirectional Long Short Term Memory layer has utilized that context this highlights over this domain consequences information; and next that multi head self-attention layer has utilized on measures comparable position over these contexts domain information. Lastly, the DGA domain features that miscellaneous weights are set within this output layer through improve detection in addition to classification tasks. Toward realistic, the captured of F1 measure 98.95% to the extent of detection together with macro-average F1 measure 83.97% to the extent of classification domain of DGA tasks.