Detecting Financial Fraud Using Rule Ensembles on Functional Covariates

• Activity Number: 356
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): David G. Whiting*+
• Companies: Brigham Young University
• Keywords: Functional Data Analysis ; Rule Ensembles ; Data Mining ; Financial Fraud

Abstract

Discovery of financial fraud has profound social consequences. Loss of shareholder value, bankruptcy, and loss of confidence in professional audit firms have resulted from failure to detect financial fraud. Recently developed ensemble methods for mining data have shown marked improvement over previous efforts in detecting financial fraud. However, used "out-of-the-box" these approaches ignore the temporal nature of many available financial data series. When observations are available over time, we use a functional data approach to register the covariates as curves using a B-spline basis. The coefficients of these covariates are used as the independent variables in a RuleFit rule ensemble model. The resulting rules, based on the B-spline coefficients, are amenable to interpretation while retaining the power and flexibility of the ensemble model.