EVALUATION OF MOVING TARGET DOH SERVER DETECTION USING MACHINE LEARNING MODELS

Abstract

The rapid adoption of DNS over HTTPS (DoH) has introduced significant challenges in balancing privacy, security, and resistance to censorship. This thesis explores the feasibility of developing a censorship-resistant DoH service, named NinjaDoH, which leverages hyperscalers and the InterPlanetary File System (IPFS) to enhance accessibility and resilience against censorship efforts. The study investigates two core research questions: first, the development and implementation of NinjaDoH , a dynamic, censorship-resistant DoH service utilizing hyperscalers and IPFS; and second, the efficiency of existing firewall solutions in targeting NinjaDoH, a moving DoH service, using various machine learning models. By evaluating firewall responses and the success of advanced machine learning techniques in identifying NinjaDoH traffic, this research highlights the strengths and weaknesses of current detection methods. The findings demonstrate the potential of IPFS as a robust, censorship-resistant solution for secure DoH communication, offering a novel framework for safeguarding internet access against state-level and organizational censorship.