Enhancing Visual Exploration of Illicit Bitcoin Transactions with Machine Learning

Abstract

Cryptocurrencies such as Bitcoin, Ethereum, and many more use blockchain technology to facilitate a decentralized, secure, and anonymous network of financial transactions. Although desirable for legal (licit) activities, these characteristics make defining and locating patterns that potentially signal illegal (illicit) activities in an increasing number of transactions difficult. Adding to this difficulty is that transaction interactions evolve over time and may be classified as neither licit nor illicit. There has been much effort in applying both machine learning and interactive visualization to a range of cryptocurrency applications. Unfortunately, the efforts for analyzing illicit cryptocurrency transactions have mostly occurred separately, neglecting the potential of combining machine learning and interactive visualization. In this paper, we present a complete, integrated pipeline and system designed for discovering and exploring illicit transactions in a Bitcoin data set. Specifically, we use machine learning to define locations of interest by classifying all unknown transactions as either licit or illicit and then pass the fully classified data set as input to a novel, interactive visualization system. Our system’s effectiveness is illustrated with an extensive use case showing how illicit transactions can be identified and their connections explored.