Transcript A RESEARCH TAxONOMY: THE APPLICATION OF DATA

8th Biennial Symposium on
Information Integrity and Information Systems Assurance
October, 2013
A RESEARCH TAXONOMY:
THE APPLICATION OF DATA MINING
TO FRAUD DETECTION
Glen L. Gray
California State University at Northridge
Roger Debreceny
University of Hawai‘i at Mānoa
Introduction
• Observation
• The application of data mining to fraud detection during
financial audits is at an early stage of development and
researchers take a scatter-shot approach
• Objectives of our study
• Explore the application of data mining techniques to fraud
detection
• Develop a taxonomy to support and guide future research
Predictive Power
Data Examination Tools
Software sophistication
GRAB THE
LOW HANGING
FRUIT ..
Return on investment in data mining
Spread investment in data
mining over many clients
Return on investment in data mining
Spread investment in data
mining for one client over
many possible fraud objects
LOOKING FOR THE
SWEET SPOT ..
Where can we leverage value from data mining in
fraud detection?
Looking for the sweet spot!
Scheme Scheme Scheme Scheme Scheme
Fraud
Fraud
Fraud
Fraud
Fraud
Fraud
Looking for the sweet spot!
Scheme Scheme Scheme Scheme Scheme
Fraud
Fraud
Fraud
Fraud
Fraud
Fraud
Looking for the sweet spot!
Scheme Scheme Scheme Scheme Scheme
Fraud



Fraud
Fraud

Fraud
Fraud
Fraud





Fraud Class by Evidence Scheme
Gao, L., and R. P.
Srivastava. 2011. The
Decomposition of
Management Fraud
Schemes: Analyses and
Implications. Indian
Accounting Review 15
(1):1-23.
Audit Specific Data Mining Scoring Scheme
Scoring Elements
Source
Target
Signals
Data Types
Semantics
Reversal Accounting Entities
Spreading of Fraudulent Items
among Accounts
Related Parties
Fake Products
Client Misrepresentations
Hidden Documents
Altered Documents
Collusion with third Parties
Fake Documents
Fraud Class, Evidence Scheme and Data
Mining
Fictitious Revenue
Premature Revenue
Recognition
Frequency
Data Mining
Applicability
Overstated
Assets/Understated
Liabilities
Nil
Low
Fictitious Assets
Low
Mediu
m
Other Measures to
Overstate Revenue
Mediu
m
High
Overvalued
Assets/Equity
High
Omitted Disclosure
FUTURE RESEARCH
Themes in Data Mining
• Mining External Information as Part of the Planning
•
•
•
•
Phase
Mining Client Non-financial performance data
Analysis of Journal Entries
Mining Accounting Information Systems
Email and other textual sources