Sequence Matching for Suspicious Activity Detection in Anti-Money Laundering

Developing effective suspicious activity detection methods has become an increasingly critical problem for governments and financial institutions in their efforts to fight money laundering. Previous anti-money laundering (AML) systems were mostly rule-based systems which suffered from low efficiency and could can be easily learned and evaded by money launders. Recently researchers have begun to use machine learning methods to solve the suspicious activity detection problem. However nearly all these methods focus on detecting suspicious activities on accounts or individual level. In this paper we propose a sequence matching based algorithm to identify suspicious sequences in transactions. Our method aims to pick out suspicious transaction sequences using two kinds of information as reference sequences: 1) individual account's transaction history and 2) transaction information from other accounts in a peer group. By introducing the reference sequences, we can combat those who want to evade regulations by simply learning and adapting reporting criteria, and easily detect suspicious patterns. The initial results show that our approach is highly accurate.