Significant Digits: Using Large-Scale Blockchain Data to Predict Fraudulent Addresses

TL;DR

This paper presents a method that leverages large-scale blockchain data, graph mining, and Benford's Law to predict fraudulent cryptocurrency addresses with reasonable accuracy.

Contribution

It introduces a novel approach combining graph mining and Benford's Law features for fraud detection in blockchain networks.

Findings

Benford's Law features are highly significant for prediction.

The gradient-boosting model achieves reasonable fraud detection accuracy.

Graph-based transaction features improve prediction performance.

Abstract

Blockchain systems and cryptocurrencies have exploded in popularity over the past decade, and with this growing user base, the number of cryptocurrency scams has also surged. Given the graphical structure of blockchain networks and the abundance of data generated on these networks, we use graph mining techniques to extract essential information on transactions and apply Benford's Law to extract distributional information on address transactions. We then apply a gradient-boosting tree model to predict fraudulent addresses. Our results show that our method can detect scams with reasonable accuracy and that the features generated based on Benford's Law are the most significant features.