Detecting Domain Generation Algorithms to prevent DDoS attacks using Deep Learning

Abstract

Denial-of-Service (DoS) and Distributed-Denial-of-Service (DDoS) attacks are some of the most destructive attacks which are leveraged to devastating effects, bringing down some of the biggest and well-known web domains and infrastructures hosted on the Internet. Attacks have been increasing in magnitude every year, peaking at a magnitude of 1.5 Terabytes per second in 2018. Domain Generation Algorithms are used by Command-and-Control servers of botnets to upload malware on certain domains, which are contacted by the bots to receive it. Thousands of domain names can be created using algorithms, which makes manual detection and sink-holing of the domains difficult. Present day malware authors use clever techniques to generate domain names which are quite similar to the authentic names which we are familiar with. Detecting these algorithms, and their respective servers effectively will lead to nipping the problem of DDoS attacks in the bud, as the server will not be able to communicate with the bots. This renders the attack ineffective. The aim of this work is to develop effective models which can detect such malware or bots, using machine learning and deep learning techniques.