（2007）AUDITING TRANSACTION DATABASES： A HYBRID APPROACH.docVIP

AUDITING TRANSACTION DATABASES: A HYBRID APPROACH Name: Argyris Argyrou University: Swedish School of Economics and Business Administration Address: Arkadiankatu 22, 00100 – Helsinki, FINLAND Telephone: + 358 – 9 – 431 – 33429 E-mail: argyris.argyrou@hanken.fi Supervisors: Bo-G?ran Ekholm and Anders Tallberg bo-goran.ekholm@hanken.fi anders.tallberg@hanken.fi 16th EDAMBA Summer Academy Soreze, France July 2007 Abstract: The paper poses the question how an auditor can audit a transaction database in situations where he/she has only minimal expert-domain knowledge and no prior information about the transactions updating the database. To this end, the study first builds and then evaluates a hybrid artefact based on self-organizing maps (SOM), inductive learning algorithm and fuzzy set theory. The study follows the design research methodology. The preliminary data consist of 1,543 general-ledger transactions, each of which is described by six attributes. Preliminary results suggest that the artefact is robust to Type I error and enjoys a high overall sensitivity rate. Keywords: Transaction database audit, self-organizing map, FS-ID3, inductive learning, Sugeno-type fuzzy inference system, design research methodology. CHAPTER 1: INTRODUCTION 1.1 Research question The study poses the question how an auditor can audit a transaction database in situations where he/she has only minimal expert-domain knowledge and no prior information about the transactions updating the database. To address this issue, the study builds and then evaluates a hybrid artefact based on self-organizing maps (i.e. SOM) (Kohonen, 2001), inductive learning and fuzzy set theory (Zadeh, 1965). In the context of this study, the term audit refers to any procedure an auditor undertakes to identify transactions that deviate form the ‘normal’ state of the database as ‘abnormal’. The study’s proposed hybrid artefa