Revealing the Character of Nodes in a Blockchain with Supervised Learning

Abstract

The term blockchain has its roots in cryptocurrencies. However, its applications are now more widespread, and in many areas, this technology has become the foundation of the distributed ledger. The blockchain protocol assumes that all the participants of the system are both contributors and safeguards of this ledger, since the lack of a trusted third party requires other security precautions in order to maintain the consistency of transactions. In this work, we investigate whether for the participants of a blockchain-based system that does not require revealing the character explicitly, it can be discovered by other means. In order to verify this, we built and publicly released a dataset of nearly 9,000 addresses of nodes in the most popular cryptocurrency - Bitcoin, and then labelled them. These labels represent the character the nodes have in the network, e.g. miners or exchanges. We then developed a set of features that quantify the behaviour of nodes in the network and used supervised machine learning algorithms to find out whether the character of nodes can be revealed based on these features. Our results demonstrate, due to the F-score reaching over 95% in the best-performing algorithms, that it is hard to hide the role the node has in a blockchain-based network. These results indicate that to build trustworthy blockchain-based systems that fully comply with original blockchain assumptions, specific countermeasures are needed in order to preserve the desired level of anonymity.