The Challenge
Open Banking, and in particular the application of international regulations such as PSD-2 since 2019, has given an edge to the banking sector by allowing the offer of more agile and innovative payment services. In this context, the request of the Banks’ Anti-Fraud Services is to improve the prevention of fraud on bank transactions in an increasingly digital and fast process.
The Solution
The solution is a Supervised Model Production (Learning) service with MLOPS services in production. Based on the transaction history, models with multiple learning processes are initialized in order to identify models with better accuracy in response and that respond to the need to lower false positives according to a “cost function” foreseen and agreed with the Customer during the Requirements phase. Specific models are also developed to respond to specific controls required by PSD-2 during transaction authorization.
Machine Learning Models are queried in real time and provide a risk score on the individual transaction. Finally, with the MLOPS service, the Models are monitored to re-train when abnormal conditions or sub-threshold scoring occur that determine the need for updating.
The Benefits
The Machine Learning Anti-Fraud Models created allow you to block transactions in real time that present anomalies already in the first phase of the payment authorization process.
The monitoring of the results of the Models allows to understand the false positives in order to reduce them and at the same time to facilitate and reduce the work of the bank’s anti-fraud team in the process of identifying anomalies and blocking possible transactions.
Humanativa’s Role
Implementation of the end-to-end Machine Learning service, from the creation of Machine Learning Models, Integration into Business Systems, MLOPS Procedures (with ApiRest & microservices), and Advanced Analytics.