Network Embedding Analysis for Anti-Money Laundering Detection

Abstract

We employ network embedding to detect money laundering in financial transaction networks. Using real anonymized banking data, we model over one million accounts as a directed graph and use it to refine previously detected suspicious cycles with node2vec embeddings, creating a new network parameter, the spread number. Combined with more traditional centrality measures, these define an aggregate score $R$ that highlights so-called anti-central nodes: accounts that are structurally important yet organized to avoid detection. Our results show only a small subset of cycles attain high $R$ values, flagging concentrated groups of suspicious accounts. Our approach demonstrates the potential of embedding-based network analysis to expose laundering strategies that evade traditional graph centrality measures.