부산대학교 도서관

Domain Name Service (DNS) has become an essential element of the Internet in current era. The traditional DNS extends its authority down to the Top Level Domains (TLDs) and individual names before returning to the Root Server. To avoid the cons related to centralized authority scheme of the traditional DNS, recently many techniques based on the decentralized Internet with support of Blockchain (BC) are emerging. Hence, currently a variety of protocols and mechanisms are focusing on implementation of the Decentralized Naming Services which in turn supports concept of decentralized internet services. These Decentralized Naming Services are totally based on Blockchain architectures and provides various pros like resistance to single point of failure, resistance to censorship and permanent proof of ownership. Blockchain Domain Names act as Digital Identity in Decentralized Internet. Security is major concern in any DNS system and the similar challenges exists in the Blockchain DNS system too. Further, there is a lot of diversity being observed in decentralized internet space as a Domain name in a Blockchain DNS can refer to a website/portal or a wallet or a Non-Fungible Token (NFT) or simply be parked. Also, currently, there is rampant growth in development of decentralized applications with no standards implemented - neither across Blockchains nor regions. Hence, there is a need to detect security exploits carried out using malicious domains in Blockchain DNS system and implement mitigation measures accordingly. In this research work, we study and analyze malicious domain names in decentralized BC system. We explored Ethereum transaction dataset for detecting malicious transactions using Machine Learning (ML) and Deep Learning (DL) algorithms with good accuracy of 99% and 95% using Decision Tree and 1 Dimensional Convolutional Neural Network (CNN) respectively. We also applied ML techniques on Ethereum wallet dark list dataset for detection of domain names that were vii used for phishing attacks. We received 96% accuracy after handling data imbalance by Random Forest model. Model predicted nature of domain based on 17 features that were extracted from Ethereum wallet dark list dataset like number of consonants, vowels, digits, symbols and their ratios to length of domain.