Exchange Pattern Mining In The Bitcoin Transaction Directed Hypergraph

Stephen Ranshous
Cliff A Joslyn
Sean Kreyling
Kathleen Nowak
Nagiza F Samatova
Curtis L West
Samuel Winters
Bitcoin exchanges operate between digital and fiat currency networks, thus providing an opportunity to connect real-world identities to pseudonymous addresses, an important task for anti-money laundering efforts. We seek to characterize, understand, and identify patterns centered around exchanges in the context of a directed hypergraph model for Bitcoin transactions. We introduce the idea of motifs in directed hypergraphs, considering a particular 2-motif as a potential laundering pattern. We identify distinct statistical properties of exchange addresses related to the acquisition and spending of bitcoin. We then leverage this to build classification models to learn a set of discriminating features, and are able to predict if an address is owned by an exchange with > 80% accuracy using purely structural features of the graph. Applying this classifier to the 2-motif patterns reveals a preponderance of interexchange activity, while not necessarily significant laundering patterns.

Metadata

Year 2017
Peer Reviewed done
Venue Financial Cryptography
mode_edit