A Scan Statistics Based Suspicious Transactions Detection Model for Anti-money Laundering (AML) in Financial Institutions

Developing effective suspicious activity detection models has drawn more and more interests for supervision agencies and financial institutions in their efforts to combat money laundering. Most previous AML systems were mainly rule-based which suffered from low efficiency and also could be easily learned and evaded by money launders. While most machine learning models for AML were focused on individual level. Our paper proposes a suspicious activity recognition method basing on scan statistics, it aims to identify suspicious sequences on transaction level for financial institutions. In the end, we evaluate our algorithm using real financial data from commercial banks. And the initial experiment results demonstrate the efficiency of our approach.