UNMASKING USER IDENTITIES BY CLUSTERING ADDRESSES AND ANALYZING BEHAVIOR RELATED TO BLOCKCHAIN TRANSACTIONS

Abstract

An explosion of online transactions has occurred with recent developments in blockchain technology and the availability of Bitcoin and other cryptocurrencies as a form of stored value. With its feature of pseudo-anonymity, cryptocurrency enables illicit transactions in which users attempt to hide their identities through various methods such as “mixers.” Meanwhile, scholarly research on how to track down address ownership based on blockchain transaction flow and other offline methods is growing, even as cyber criminals exploit more advanced methods to obfuscate their identities and companies working with law enforcement seek to counter them. This research aims to help fight illegal activities by contributing to the identification of criminals who use blockchain technology for illicit transactions. In order to validate our hypothesis that specific, discernible transaction patterns exist among different users, we demonstrate clustering of blockchain addresses using features derived from users’ temporal behavior in making their transactions. The research shows promising results in clustering addresses to their related entities. We also use this clustering model to induce a predictive model of identity. Using a 1-year dataset, our model finds the mutual information between wallets and clusters to be 4.08 bits. Overall, our validation result shows that our model is able to predict wallet based entirely on transaction behavior at 384 times greater than chance accuracy.