Association Rules for Fraud Detection

Abstract

A recent study reported that businesses normally lose 5% of their revenues to fraud each year—with the median loss for occupational fraud set at $140,000. But finding fraud is tough: 87% of the perpetrators have never been previously charged with a fraud‐related offense. And budgetary cutbacks are another obstacle for internal auditors, who are continuously being asked to do more with less. Computerized records offer greater accuracy and efficiency, but investigators can find themselves drowning in a sea of digital data. So fraud investigators need a practical way of finding the proverbial “needle in a haystack” clue that will lead them to the perpetrators. The author suggests one solution: association rules (AR), a data‐mining technique that can help auditors find fraud when they are not sure where to search. He explains how auditors can use AR to find patterns and relationships that would otherwise go unnoticed. © 2013 Wiley Periodicals, Inc.

Data‐driven auditing: A predictive modeling approach to fraud detection and classification

“…In this regard, anomaly detection methods like unsupervised outlier detection could be considered (Campos et al, ); however, for the course of this study, we decided to employ more salient indicators as discussed in the audit literature. Exhibit presents an overview of the indicators for suspicious transactions and possible underlying management frauds (Singh et al, ; Tackett, ).…”

The use of predictive analytics in finance

“…Association rules (AR) use Machine Learning (ML) models to check for patterns or related events in a database. For example, they are suggested by 26 as a way of detecting fraud in banks. Transactions or events that are usually associated with fraud, such as claiming frequent refunds, can be data mined and then flagged.…”

Is Benford’s Law Effective in Fraud Detection for Expense Cycle?

“…Bella et al 25 developed a four-stage self-organizing map fraud detection architecture of electronic billing records. Tackett 26 suggested the use of association rules in detecting fraud through finding patterns and relationships when examining a company's digital records. Barney and Schulzke 27 underline how reliable is the BL in audit testing as false positives test results (Type I errors) show that the digital analysis is costly and unnecessary.…”