Nida S. Khan, Asma S. Larik, Quratulain Rajput, and Sajjad Haider
Anti-money laundering, Bayesian network, suspicious activityreporting
This paper presents a Bayesian network (BN)-based approach to analyse customers’ transactions in a financial institution and then to detect suspicious patterns in them. The approach is developed as part of an anti-money laundering system that requires identification of suspicious transactions so that they are reported to the concerned authorities in a timely manner. The proposed BN is designed on the basis of rules suggested by the State Bank of Pakistan in its 2008 regulations to declare a transaction as suspicious. Using transaction history, the proposed approach assigns a baseline money laundering score to each customer. The score is an indication of the customer’s transaction behaviour. During the live operational mode, if there is a significant difference in customer’s historical transactional pattern and the current behaviour, an alert is generated which requires the branch manager (or compliance head) to verify the reason for the difference. The approach has been tested on real financial transactions set having more than 8.2 million records of more than hundred thousand customers.
Important Links:
Go Back
