CubeFlow: Money Laundering Detection with Coupled Tensors

Money laundering (ML) is the behavior to conceal source of money achieved by illegitimate activities, and always be a fast process involving frequent chained transactions. How can we detect ML fraudulent activity in large scale attributed transaction data (i.e. tensors)? Most existing methods dense blocks graph or tensor, which do not consider fact that are frequently transferred through middle accounts. CubeFlow proposed this paper scalable, flow-based approach spot fraud from mass transactions modeling them as two coupled tensors applying novel multi-attribute metric reveal transfer chains accurately. Extensive experiments show outperforms state-of-the-art baselines detection both synthetic real data.