Exploration of Hybrid Neural Networks for Domain Name Generation

The increase in the use of Domain Generation Algorithms (DGA) for communication between Command and Control (C &C) servers presents significant challenges in botnet detection. This study investigates the use of a hybrid neural network architecture that combines Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) layers for algorithmic domain name generation. The purpose of this domain name generator is to increase the availability of algorithmically generated domains (AGD), which can be used in future research and contribute to improving the capabilities of AGD detectors in upcoming investigations. To validate the effectiveness of the generator, three evaluation methods were employed. First, the registration status of the generated domains was evaluated using a Domain Availability Verification (DAV) tool, revealing that 93% of the domains were not registered. Second, the generated domains underwent Plausibility Assessment (PA) using a large-scale language model (LLM), which classified only 8% as suspicious. Finally, an Algorithmic Detection Effectiveness (ADE) evaluation was conducted using AGD detectors from the literature, demonstrating low detection rates of 3% and 16%, respectively. These results highlight the capability of the studied model to generate domain names similar to legitimate ones and underscore the need to enhance AGD detection systems.