Breaking Bad: De-Anonymising Entity Types on the Bitcoin Blockchain Using Supervised Machine Learning

Abstract

Bitcoin is a cryptocurrency whose transactions are recorded on a distributed, openly accessible ledger. On the Bitcoin Blockchain, an entity's real-world identity is hidden behind a pseudonym, a so-called address. Therefore, Bitcoin is widely assumed to provide a high degree of anonymity, which is a driver for its frequent use for illicit activities. This paper presents a novel approach for reducing the anonymity of the Bitcoin Blockchain by using Supervised Machine Learning to predict the type of yet-unidentified entities. We utilised a sample of 434 entities (with ≈ 200 million transactions), whose identity and type had been revealed, as training set data and built classifiers differentiating among 10 categories. Our main finding is that we can indeed predict the type of a yet-unidentified entity. Using the Gradient Boosting algorithm, we achieve an accuracy of 77% and F1-score of ≈ 0.75. We discuss our novel approach of Supervised Machine Learning for uncovering Bitcoin Blockchain anonymity and its potential applications to forensics and financial compliance and its societal implications, outline study limitations and propose future research directions.

An Evaluation of Bitcoin Address Classification based on Transaction History Summarization

“…To deal with imbalanced data, sampling-based approach and cost-sensitive approach are considered simultaneously in [18]. To reduce anonymity of Bitcoin by predicting yet-unidentified addresses, [19] trained classifiers with synthetic minority over-sampling technique [20] on imbalanced data. [21] introduces the idea of motifs in directed hypergraphs, defining exchange patterns of addresses.…”

“…We leverage the dataset collected by [5] in order to facilitate the comparison between our method and the previous work. As described in Table II, the dataset contains totally 26,313 addresses with labels and owners, which are derived from a simple heuristic, naming multi-input transactions [11], sharedsend clustering [12], co-spend clustering [19], or common spending [6]. The idea is that the addresses of inputs in a transaction belong to the same entity because spending bitcoins needs the signature of the owner's private key.…”

Bitcoin and Cybersecurity: Temporal Dissection of Blockchain Data to Unveil Changes in Entity Behavioral Patterns

“…In [10,12], methods for attacking Bitcoin user anonymity are presented. Both methods use the whole blockchain to create supervised machine learning models and classify Bitcoin entities.…”

“…Prior studies have tried to classify entities according to classes representing specific entity behavior within the network [10][11][12]. These techniques usually consider the whole blockchain and thus classification is performed considering all network (macro) dynamics among users.…”

Blockchain-based Personal Health Data Sharing System Using Cloud Storage

“…The blockchain technology has gained substantial popularity in recent years, primarily in financial field, due to the cryptocurrencies. For example, Bitcoin was first introduced in 2008 [7] and ever since has attracted the attention of the research community from diverse academic fields [8], [9], [10] and gained mainstream popularity due to its unique characteristics, such as the absence of centralised control, an assumed high degree of anonymity and distributed consensus over decentralised networks. Blockchain solutions could reduce data breach risks by utilising threshold encryption of data together using public key infrastructure, where cooperation of multiple parties is required to decrypt data and asymmetric cryptography is used to authenticate communication with system participants [11].…”