Exploration and Exploitation in Deciding What to Audit

• In Event: Information Utility

Abstract

There are several challenges in applying analytical models to the auditing problem
of identifying irregular transactions. Specically, the auditor receives only one-sided
feedback, i.e. she learns the true nature of only those transactions that were investi-
gated previously, and has a limit on the number of transactions that can be investigated
(due to limited audit resources). We argue that because of these challenges standard
statistical models may not be well-suited for auditing and have to be modied in order
to achieve better performance. In this paper we propose a framework to boost the per-
formance of the statistical models in auditing. We show an example how to apply this
framework to some popular statistical models. The results of testing the framework
on the real-world data show a signicant increase of performance for some statistical
models.

Authors

• Roman Chychyla, Rutgers, The State University of New Jersey

• Alexander Kogan, Rutgers, The State University of New Jersey, Newark