Bitcoin Mixing Detection Using Deep Autoencoder

Abstract

Bitcoin is a decentralized transaction platform and the largest cryptocurrency system. Bitcoin represents a chain of blocks containing its entire legal transaction history, thereby providing convenience for tracking money. However, mixing services are used as an effective means to hide the identity of a transaction address by combining several transfers from different users. Detecting the original user of a Bitcoin address and the money flow is essential in some special circumstances such as anomaly detection. Recognizing Bitcoin mixing services and de-mixing user accounts have only rarely been studied. Here we demonstrate that Bitcoin transaction graphs possess community properties and that a mixing service can be regarded as a cluster outlier. Motivated by the success of graph embedding in social network analysis, we propose a feature-based method to identify mixing services, testing our method on the real Bitcoin ledger.