Cluster Analysis for Anomaly Detection in Accounting Data: An Audit Approach

Abstract

This study examines the application of cluster analysis in the accounting domain, particularly discrepancy detection in audit. Cluster analysis groups data so that points within a single group or cluster are similar to one another and distinct from points in other clusters. Clustering has been shown to be a good candidate for anomaly detection. The purpose of this study is to examine the use of clustering technology to automate fraud filtering during an audit. We use cluster analysis to help auditors focus their efforts when evaluating group life insurance claims. Claims with similar characteristics have been grouped together and small-population clusters have been flagged for further investigation. Some dominant characteristics of those clusters which have been flagged are large beneficiary payment, large interest payment amounts, and long lag between submission and payment.

A Novelty Detection Algorithm for Multiple Normal Classes : Application to TFT-LCD Processes

“…이상치 탐지(novelty detection)는 하나의 정상집단을 특성화하 는 모델을 학습시킨 후 새로운 관측치와의 비교를 통해 새로 운 관측치의 비정상 여부를 확인하는 기계 학습 기법으로 (Kang et al, 2012) 사기 탐지 (Issa and Miklos, 2011;Thiprungsri and Miklos, 2011), 침입 탐지 (Kumar et al, 2005), 의료 (Tarassenko et al, 2006), 제조공정 과정 (Clifton et al, 2008) 등 다양한 분야에 서 응용되고 있다. 특히 제조 공정 과정에서 불량 제품으로 인한 재작업 증가는 효율성을 감소시키는 주요 원인으로 TFT-LCD 나 PCB와 같은 하이테크 산업에서 이상치 탐지는 중요한 문제 로 인식되고 있다 (Shin, 2010).…”

Internal Control Computerization for Derivatives

“…When management cannot determine appropriate thresholds, one solution is to call upon the unsupervised learning algorithm of cluster analysis to identify abnormal transactions. Thiprungsri and Vasarhelyi (2011) propose using cluster analysis to help auditors detect fraud, and report on applying cluster analysis to pinpoint anomalies in group life insurance claims. Therefore, the computerization of internal controls illustrated in this paper can be integrated with cluster analysis to supplement management's judgment.…”

AIS in an Age of Big Data

“…The context in which this text is presented must be inferred by the type of document on which it is found, titles, subheadings, and words within the text itself. Major themes can be extracted using mathematical and machine learning techniques such as tf-idf (Aizawa 2003), latent semantic analysis (Landauer et al 1998), and cluster analysis (Thiprungsri and Vasarhelyi 2011). Data with free text can be ''tagged'' based on the context, but not with the granularity and accuracy of XBRL.…”