AI and Data Science Master the art of data science. Join now
Banks are facing an increasing demand from regulators to monitor and mitigate money laundering risks. One common challenge is that Anti-Money Laundering (AML) Transaction Monitoring systems do not always result in suspicious investigations or actionable results. These are due to an increase of false-positive alerts within financial institutions.
By understanding and exploring the business process in the context of the data, it is easier to develop a successful approach to identify suspicious activity and understand what defines a ‘normal’ activity. Not all captured activities are important indicators of suspicious activity and activities differs over time. Therefore, banks needs to revisit thresholds of suspicious activity and perform threshold tuning. This needs to be a continuous process of model tuning and improvement.
This article discuss a top-down and bottom-up approach of customer segmentation that can be applied to AML Transaction Monitoring systems. The top-down methodology is derived from business knowledge and known customer attributes. The bottom-up approach, on the other hand is a data driven methodology involving data mining and unsupervised modelling methods used to identify homogenous customer groups with similar transactional behaviour.