The Process of Integrating Machine Learning into Risk Based Authentication
A dataset that contains over 25 million records a day and potential for over 1000 aggregates makes any machine learning effort challenge. Now add in supplemental data sources. Now add-in a target rate below 1%; now add in incomplete information about your target. Now imagine developing this to be the first real-time, interdictive machine learning model in an environment that has high customer impact. The risk of analysis paralysis is present and dangerous.
These were the challenges faced when developing the Fraudulent Authentication Model. However, through a combination of adopting agile methodology and analytical decision making we were able to overcome the risks from these challenges. We were able to use segmentation to simultaneously reduce the challenges from the dataset size and the customer impact. We were able to take the incomplete nature of fraud targets and turn it into an asset in our machine learning journey. In the end, we were able to turn our most complex project into our most successful machine learning model!
Matt Hall is a Data Scientist on the Data Science Delivery team within USAA Federal Savings Bank’s Fraud and Central Operations organization. Matt Hall leads a variety of machine learning efforts aimed at preventing fraudsters from accessing members’ accounts. Matt Hall was the head developer on FCO’s interdictive solutions based on machine learning models. In addition to his analytics roles, Matt Hall leads the internship program for the Data Science Delivery team.
Prior to USAA, Matt Hall worked as a Data Analyst in the Department of Student Life’s Assessment and Planning office.
Matt Hall is a two-time graduate of the University of Alabama, earning his M.S. in Applied Statistics and B.S. in Economics
stay up to date
Sign up for email updates on the 2022 symposium, and other analytics news within the Culverhouse College of Business.
This website uses cookies to collect information to improve your browsing experience. Please review our Privacy Statement for more information.
