A Classification Method for Blockchain Addresses Considering Features of Address Clusters
Tomohiro Ojika, Shin Morishima · 2024
Blockchain is widely used for various applications due to its different characteristics compared to traditional financial networks such as banking systems. However, its anonymity also facilitates illegal money flows such as money laundering. Address classification methods have been proposed to prevent these illegal activities by tracking the flow of funds. In blockchain, an address cluster, which consists of a set of addresses created by the same user, represents a substantial trading entity. However, existing methods typically classify addresses based on individual address features without considering address clusters. In this paper, we propose an address classification method that considers address clusters and utilizes the internal features of these clusters. In addition, to further improve the classification accuracy by enhancing the estimation of address clusters, we propose an address clustering method based on the characteristics of exchanges, which are influential users. Through evaluations comparing the proposed and existing methods, our approach demonstrates a significant improvement in accuracy. Specifically, the proposed method achieves a 21.7% increase in accuracy in 5-category classification and a 32.2% increase in 2-category classification, distinguishing between exchange and non-exchange categories.