Abstract: Machine learning techniques applied to business problems are often subject to the criticism that they are ""black boxes"" which give decisions without providing to users any insight into how those decisions were arrived at. Such insight into the reasons behind decisions can help guide users to make the best use of decisions from machine learning algorithms; in some cases understanding/justification of decisions is required by regulators.
AdaBoost is one of the most successful machine learning algorithms in a wide variety of application areas including the financial services industry. In this talk we will review the AdaBoost algorithm and present a method, "influence diagrams," for displaying the ""thinking"" behind AdaBoost decision-making. We will illustrate the use of influence diagrams applied to financial-fraud detection and discuss how this technique, although developed for the purpose of communicating information to non-technical users, can be employed for tuning the AdaBoost system.
Bio: Mark Rubin is Chief Data Scientist at FIS. He has worked in theoretical high-energy physics at the University of Texas at Austin, Fermilab and Rockefeller University, and in machine learning and machine vision at the Naval Air Warfare Center at China Lake, Boston University and MIT Lincoln Laboratory. His current activities include applications of machine learning to financial fraud detection and quantitative studies in foundations of quantum mechanics.
Mark Rubin, PhD
Data Science Team Lead at FIS