Journal entry anomaly detection model

• Author: Svjetlana Letinic,Verica Budimir,Mario Zupan
• Published date: 01 October 2020
• DOI: http://doi.org/10.1002/isaf.1485
• Date: 01 October 2020

Received: 19 February 2020 Revised: 8 November 2020 Acce pted: 8 November 2020

DOI: 10.1002/isaf.1485

RESEARCH ARTICLE

Journal entry anomaly detection model

Mario Zupan Verica Budimir Svjetlana Letinic

Social Department,Polytechnic in Pozega,

Pozega, Croatia

Correspo ndenc e

Mario Zupan, So cial Department, Polytec hnic

in Pozega, Pozega, Croatia.

Email: mzupan@vup.hr

Summary

Although numerous scientific papers have been written on deep learning, very few

have been written o n the exploitation o f such tech nology in the f ield of acco unting

or bookkee ping. O ur scientific study is o riented exac tly toward this specific field.

As accountants, we know the problems faced in modern accounting. Although

accountants may have a plethora of information regarding technology support,

looking for errors or fraud is a demanding and time-consuming task that depends on

manu al skills a nd pro fes sion al know ledg e. O ur eff orts are o riente d tow ard reso lvin g

the pro blem o f error- det ecti on aut omati on tha t is curre ntly po ssib le thro ugh n ew

technologies, and we are trying to develop a web application that will alleviate the

problems of journal entry anomaly detection. Our developed application accepts

data from one specific enterprise resource planning system while also representing

a general software framework for other enterprise resource planning developers.

Our web application is a prototype that uses two of the most popular deep-learning

architectures; namely, a variational autoencoder and long short-term memory. The

application was teste d on two diffe rent journals: data set D, learned on acc ounting

journals from 2007 to 2018 and then tested during the year 2019, and data set H,

learned on journals from 2014 to 2016 and then tested during the year 2017. Both

acco unting jou rnals were ge nerated by m icro en trepreneu rs.

KEYWORDS

accounting control system,anomaly detection, bookkeeping, deep learning, generalledger

1INTRODUCTION

Croatian small and medium enterprises have a wide pallet of quality

enterp rise res ource p lanni ng (ERP) sy stem s deve loped b y nume rous

domestic software companies. As in everyinfo rmationsystem, human

effort, as well as interaction between modules, can cause errors.

Anomalies in acc ounting boo ks occur on a daily bas is, and unin-

tentional human errors, attempted fraud, and continuous legislative

chan ges are so me of th e critic al caus es. V ery rare acc oun ting tran s-

actions recorded we will call anomalies. The debit side of the equity

account recorded by a VAT document, for instance, is not a stan-

dard procedureand needs to be red-flagged. Anomalies could happen

intentionallyor unintentionally,violating accounting rules or not. Unin-

tentional anomalies occur despitethe f act thatmost existing controls

integrated into accounting modules of modern ERP systems are cre-

ated inco mpliancewith bookkeeping rules. Because smalland medium

enterp rises d o not h ave au dit ob ligatio ns reg ulate d by Croa tian law ,

manualtax inspections are the o nly mechanism of their accounting and

tax control. In gene ral, the detection of errors, made intentionally or

not, consumes a largeportion of a bookkeeper'so rtax inspector's time,

and the c orrec tion of e rrors is no t an easy p art of th eir job, p articu -

larly owing to the architectu rean df unctioning of acco unting software

modules. Namely, most of today's ERP systems have specializeddoc-

uments (digital forms) fo rev ery specific business even t. Every digital

form is connected with one or more journal entry schemes created

by senior accountants. Junior accountants or non-accountant employ-

ees do not have to be familiar with journal schemes be cause they

communicate only through forms. As long as the modern accounting

modules inside ERP systems are functioning based on the princi-

ple described, a single error in only one journal entry scheme can

cause an incorrect accounting entry for a whole set of connected

digital forms.

IntellSys Acc Fin Mgmt. 2020;27:197–209. wileyonlinelibrary.com/journal/isaf © 2020 John Wiley & Sons, Ltd. 197