In order to detect unknown patterns of money-laundering, RiskQuest developed an unsupervised machine learning model
The task at hand was to devise an unsupervised machine-learning model that could identify money laundering schemes by leveraging the transaction data of the bank’s customers. The project had a tight deadline of just three months and, once in production, the model would become the challenger of a third-party AML system already in place at the time.
RiskQuest, alongside some of the bank’s internal data scientists, developed an ensemble of unsupervised machine learning algorithms to identify anomalous behaviours in the transaction profile of the bank’s customers. The chosen models originate from very different types of machine learning frameworks, ranging from neural-networks to random forest and clustering-based techniques. Ensembling the decision process of different statistical methods in a single one typically results in a more powerful model which is able to produce very accurate predictions.
The outcomes of the unsupervised model were iteratively tested by sending generated alerts to transaction-monitoring analysts for feedback. After the initial three months development time the model was deemed sufficiently powerful to be preferred above the vendor solution. RiskQuest used their technical expertise to create a better model than the third-party model in a short amount of time.