Identity Discovery in Bitcoin Blockchain: Leveraging Transactions Metadata via Supervised Learning

Abstract

Blockchain-based systems such as the one proposed to support the Bitcoin protocol are primarily used to enable the execution of financial transactions in a decentralized manner. The characteristics of blockchains have inspired the development of new types of applications that are shifting from its original purpose. Besides supporting the recording of crypto-currency transactions blockchains are also being exploited as mediums of recording arbitrary chunks of data. One technique for embedding such data on the public Bitcoin blockchain is using the OP_RETURN opcode creating an unspendable transaction. In this paper, we leverage data retrieved from such transactions to reveal the identity of the transacting entity. In more detail, we cast the problem of identity discovery as a classification problem. An empirical evaluation using various supervised classification models (from Naive Bayes to deep learning) yield up to 99.98% classification accuracy. In addition, it is confirmed that our feature engineering methodology on using the leading characters of the OP_RETURN instruction holds a significant discrimination power when compared against the baseline.