Machine Learning-Driven Alert Optimization

This chapter discusses machine learning approach for the financial transaction monitoring. One common theme that we would recommend is to start with simplistic approach containing readily available data of good quality. Build slowly additional information like more variables or factors, network-related information, and then eventually work on transaction-level monitoring. However, to start with it could be alert scoring as it seamlessly integrates into the existing transaction monitoring for the organization. There are multiple methodologies proposed by different researchers. Even we have proposed an approach. Practitioners can adopt one of the proposed approaches or create a new one. Our proposed approach creates a customer view as it provides multidimensional view of the risk that the customer undertakes and then percolates that down to scenario level, given the type of transaction risks a particular scenario provides. A detailed overview of the model development process is explained in the appendix. With the right applicability of the framework, a user with non-technical background should also be able to work with the data scientists to lead the machine learning-driven alert scoring or transaction scoring.