Detecting anomalous behaviour using explainable AI for AML Qorus Banking Innovation Awards 2022

• Radical Approach: Moving away from the 500K cash/non-cash report philosophy which was centred on simple rule based alerts, in-house Machine Learning algorithms were used to identify gaps in the existing alert based system and detect anomalous transactional behaviour. Over and above the machine learning solution, a local explainer algorithm was implemented to interpret the alerts and aid the account officers in investigations. To this effect, a dashboard was devised at the request of some lines of business to further ease their investigations. • Digital Transformation: Prior to the project, the 500K cash/non-cash report was being generated in hard copies, causing an intensive use of paper and ink. The first digital version of the solution was made available on the Cognos platform where account officers could log-in and access the reports. Following subsequent improvements to the report, the anomaly alerts have now been centralised on the Financial Crime Risk Management (FCRM) platform, from which account officers receive notifications directly through email. A dedicated dashboard for tracking was implemented, reducing manual calculations at the level of Compliance SBU and enhancing analytical possibilities on AML data. • Infrastructure Advancements: The project leveraged latest developments in big data to build the first multi-threaded parallel processing of transactions and other information at MCB. Now, over 300GB of RAM is being used to analyse 500+ variables and 100+ million rows. Unified Solution: The idiosyncrasies of the different lines of business are now being catered for, as each has its own algorithm. This warranted the building of nine machine learning models in a short span of 3 months.