Gadžo, Am a; Suljić, Mi za; Jusu o ić, Adisa; Filipo ić, Slađana; Suljić, E na
A icle — Published Ve sion
Da a mining app oach in de ec ing inaccu a e inancial
s a emen s in go e nmen -owned en e p ises
C oa ian ope a ional esea ch e iew
Sugges ed Ci a ion: Gadžo, Am a; Suljić, Mi za; Jusu o ić, Adisa; Filipo ić, Slađana; Suljić, E na
(2025) : Da a mining app oach in de ec ing inaccu a e inancial s a emen s in go e nmen -owned
en e p ises, C oa ian ope a ional esea ch e iew, ISSN 1848-9931, C oa ian Socie y o Ope a ions
Resea ch, Zag eb, Vol. 16, Iss. 1, pp. 1-15,
h ps://doi.o g/10.17535/c o .2025.0001
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/324921
S anda d-Nu zungsbedingungen:
Die Dokumen e au EconS o dü en zu eigenen wissenscha lichen
Zwecken und zum P i a geb auch gespeiche und kopie we den.
Sie dü en die Dokumen e nich ü ö en liche ode komme zielle
Zwecke e iel äl igen, ö en lich auss ellen, ö en lich zugänglich
machen, e eiben ode ande wei ig nu zen.
So e n die Ve asse die Dokumen e un e Open-Con en -Lizenzen
(insbesonde e CC-Lizenzen) zu Ve ügung ges ell haben soll en,
gel en abweichend on diesen Nu zungsbedingungen die in de do
genann en Lizenz gewäh en Nu zungs ech e.
Te ms o use:
Documen s in EconS o may be sa ed and copied o you pe sonal
and schola ly pu poses.
You a e no o copy documen s o public o comme cial pu poses, o
exhibi he documen s publicly, o make hem publicly a ailable on he
in e ne , o o dis ibu e o o he wise use he documen s in public.
I he documen s ha e been made a ailable unde an Open Con en
Licence (especially C ea i e Commons Licences), you may exe cise
u he usage igh s as speci ied in he indica ed licence.
h ps://c ea i ecommons.o g/licenses/by-nc/4.0/
C oa ian Ope a ional Resea ch Re iew 1
CRORR 16:1(2025), 1–15
Da a mining app oach in de ec ing inaccu a e inancial s a emen s in
go e nmen -owned en e p ises
Am a Gadžo1,∗, Mi za Suljić2, Adisa Jusu o ić1, Slađana Filipo ić1and E na
Suljić3
1Facul y o Economics, Uni e si y o Tuzla, U e a Vejzagića 8, 75000 Tuzla, BiH
E-mail: ⟨{am a.gadzo, adisa.jusu o ic, sladjana. ilipo ic}@un z.ba⟩
2Cen e o Quali y Assu ance and In e nal E alua ion, Uni e si y o Tuzla, A mije RBiH bb, 75000
Tuzla, BiH
E-mail: ⟨[email p o ec ed]a⟩
3Public Elemen a y School "Simin Han", Sa ajac 4, 75207 Tuzla, BiH
E-mail: ⟨e na.c[email p o ec ed]om⟩
Abs ac . The s udy aims o assess he capabili y o a ious da a mining echniques in de ec ing
inaccu a e inancial s a emen s o go e nmen -owned en e p ises ope a ing in he Fede a ion o Bosnia
and He zego ina (FBiH). Inaccu a e inancial s a emen s indica e po en ial inancial aud. P edic ion
models o ou classi ica ion algo i hms (J48, KNN, MLP, and BayesNe ) we e examined using a da ase
comp ising 200 audi ed inancial s a emen s om go e nmen -owned en e p ises unde he supe ision
o he Audi O ice o he Ins i u ions in he Fede a ion o Bosnia and He zego ina. The esul s
ob ained h ough da a mining analysis e eal ha a da ase encompassing se en balance shee i ems
p o ides he mos comp ehensi e depic ion o inancial s a emen quali y. These se en a ibu es a e:
opening en y o accoun s ecei able, p o i (loss) a he end o he pe iod, ope a ing asse s a he
end o he pe iod, accoun s ecei able a he end o he pe iod, opening en y o ope a ing asse s,
sho e m inancial in es men s a he end o he pe iod, and opening en y o sho - e m inancial
in es men s. By employing hese se en a ibu es, he MLP algo i hm was implemen ed o cons uc
he mos p ecise p edic i e model, achie ing a 76% accu a e classi ica ion a e o inancial s a emen s.
Le e aging he iden i ied a ibu es, a ma hema ical model could po en ially be o mula ed o e ec i ely
p edic inancial s a emen s o go e nmen -owned en e p ises in FBiH. This, in u n, could conside ably
acili a e he p ocess o selec ing GOEs o inclusion in he annual wo k plan o s a e audi o s. P esen ly,
due o esou ce cons ain s, go e nmen -owned en e p ises in FBiH do no unde go egula annual
sc u iny by s a e audi o s, wi h only 10 o 15 such en e p ises being subjec o audi s each yea . The
esul s o his esea ch can also be bene icial o bo h he public and he Financial In elligence Agency in
he FBiH. The pape con ibu es o illing he gap in he li e a u e ega ding he applied me hodology,
pa icula ly in he pa conce ning he a ibu es used in he esea ch.
Keywo ds: da a mining, inancial s a emen auds, go e nmen -owned en e p ises, p edic ion o
inancial s a emen s accu acy
Recei ed: Ap il 20, 2024; accep ed: Oc obe 4, 2024; a ailable online: Feb ua y 4, 2025
DOI: 10.17535/c o .2025.0001
O iginal scien i ic pape .
∗Co esponding au ho .
This is an open access a icle unde he CC BY-NC-ND 4.0 license 1
h p://www.hdoi.h /c o -jou nal ©2025 Copy igh o his a icle is e ained by he au ho (s).
CRORR 16:1 (2025), 1–15 Gadžo e al.: Da a mining app oach in de ec ing inaccu a e inancial s a emen s...
1. In oduc ion
T ansi ional coun ies, including Bosnia and He zego ina, o en ace p oblems such as an in-
e icien public sec o , a high co up ion index, and he employmen o poli ically a ilia ed
pe sonnel in he managemen s uc u e and supe iso y boa ds o go e nmen -owned en e -
p ises. The esou ces o hese en e p ises a e o en used o suppo poli ical campaigns and
a o ce ain supplie s in he p ocu emen o goods and se ices, unde he guise o he Public
P ocu emen Law. The challenges aced by ansi ional coun ies like Bosnia and He zego ina,
including an ine icien public sec o , high co up ion index, and poli ically a ilia ed pe sonnel
in go e nmen -owned en e p ises, ha e been widely documen ed [3]. These issues ha e e oded
public us in s a e ins i u ions and led o inancial embezzlemen in go e nmen -owned en-
e p ises. An i-co up ion policies and good go e nance p ac ices ha e been implemen ed o
add ess hese challenges, bu hei e ec i eness in ebuilding us in public ins i u ions emains
a subjec o deba e [3]. The es uc u ing o s a e-owned en e p ises, wi h a ocus on co po a e
go e nance and he ole o he s a e, has been p oposed as a po en ial solu ion. Howe e , he
complex ela ionship be ween ax e asion, s a e capaci y, and us in ansi ional coun ies,
as well as he p e alence o public p ocu emen co up ion in Bosnia and He zego ina, u he
complica e he si ua ion. Go e nmen -owned en e p ises ha a e owned by he s a e o lowe
le els o go e nmen a e equen ly accused o inancial embezzlemen and do no enjoy public
us in he accu acy o hei inancial s a emen s. Financial s a emen aud in ol es he in-
en ional concealmen o omission o c i ical in o ma ion esul ing om a delibe a e ailu e o
epo inancial da a in line wi h gene ally accep ed accoun ing p inciples. Financial aud is a
se ious p oblem wo ldwide and is pa icula ly p onounced in companies ha a e s a e-owned in
coun ies such as Bosnia and He zego ina. Lalić, Jo ičić & Bošnjako ić [20] highligh s he in low
o money om ab oad and he p esen a ion o ope a ing losses as common examples o inancial
aud in he egion. Buljubašić Musano ić & Halilbego ić [2] u he unde sco es he manip-
ula ion o inancial s a emen s in ailing small and medium-sized en e p ises, wi h signi ican
di e ences in acc uals, asse quali y, le e age, p o i abili y, and liquidi y be ween ailing and
non- ailing SMEs. Isako ić-Kaplan e al. [13] explo es he applica ion o Ben o d’s Law in de-
ec ing po en ial ea nings manipula ion in income s a emen s o economic en i ies, emphasizing
he need o addi ional o ensic in es iga ions. Yadia i, Rezwiandha i & Ramdany [35] highligh
a b oade pe spec i e by iden i ying ac o s such as inancial s abili y, ex e nal p essu e, indus-
y na u e, di ec o changes, and collabo a ion wi h go e nmen p ojec s as possible indica o s
o audulen inancial epo ing in s a e-owned en e p ises. Ano he issue also a ises om he
ac ha inancial s a emen s o go e nmen -owned en e p ises a e no subjec o annual audi s
by s a e audi o s. Due o he ex ensi e numbe o ins i u ions unde s a e audi supe ision
(mo e han 2,000 ins i u ions), only a ew o go e nmen -owned en e p ises a e inco po a ed
in o he annual audi plan. The Audi O ice o he Ins i u ions in he Fede a ion o Bosnia
and He zego ina does no u ilize mode n da a mining ools o aid in he de ec ion o inaccu a e
inancial s a emen s. In he p ocess o planning which go e nmen -owned en e p ises will be
included in he audi plan o he yea , s a e audi o s a e guided by media epo s, anonymous
complain s, and in e nal in o ma ion om p e ious inancial audi s. The undamen al esea ch
ques ion in his s udy is which balance shee i ems and da a mining echnique p o ide he
bes p obabili y o p edic ing inaccu a e inancial s a emen s in go e nmen -owned en e p ises?
The e o e, he objec i es o his pape a e o iden i y he balance shee i ems ha bes indi-
ca e inaccu a e inancial s a emen s and o de e mine which da a mining echnique achie es
he bes esul s in p edic ing inaccu acies in inancial s a emen s o go e nmen -owned en e -
p ises. Da a mining echniques will be employed o de ec balance shee i ems om inancial
posi ion epo s o en e p ises ha o e he mos accu a e p edic ions o he s a e audi o s’
assessmen s o inancial s a emen quali y. Fu he mo e, a ious da a mining echniques will
be u ilized o e alua e hei p edic i e e icacy h ough compa a i e analysis o o ecas ing e-
2
CRORR 16:1 (2025), 1–15 Gadžo e al.: Da a mining app oach in de ec ing inaccu a e inancial s a emen s...
sul s. This pape aims o con ibu e o add essing he li e a u e gap conce ning he a ibu es
used o iden i y balance shee i ems ha bes p edic inaccu acies in inancial s a emen s o
go e nmen -owned en e p ises in FBiH. These objec i es will be achie ed h ough a sys ema ic
li e a u e e iew, iden i ying gaps in he li e a u e. Following his, a de ailed explana ion o
he esea ch me hodology and he applica ion o da a mining echniques will ollow. Finally,
he esea ch indings will be p esen ed, accompanied by hei explana ions.
2. Li e a u e e iew
A ange o s udies ha e explo ed he use o a ious models and echniques o p edic inac-
cu a e inancial s a emen s based on audi opinion. Gadžo e al. [5] ound ha he Beneish
M-sco e model, pa icula ly i s pa ial indica o s, can accu a ely p edic he quali y o inancial
s a emen s in public en e p ises. Simila ly, Sánchez-Se ano e al. [27] de eloped a model o
p edic ing audi opinion in consolida ed inancial s a emen s, achie ing high accu acy. Wu & Li
[33] u he imp o ed on his by using a BP neu al ne wo k wi h Adam op imize o p edic au-
di opinions in lis ed companies, wi h a high accu acy a e. Yue, Shen & Chu [37] also iden i ied
speci ic inancial a ios, such as ne asse s pe sha e and ea nings pe sha e, as s ong indica o s
o alse inancial a ai s, on he basis o he model o logis ic eg ession analysis. Resea ch con-
sis en ly shows ha inancial a ios a e a key ac o in p edic ing an audi o ’s quali ied opinion
on inancial s a emen s [6]. These a ios, such as e ained ea nings o o al asse s, equi y o
o al liabili ies, and ne income o o al asse s, a e used in a ious models o accu a ely classi y
quali ied and unquali ied opinions. Rudkhani & Jabba i [6] ound ha only wo inancial a ios,
"ea nings pe sha e" and " ixed asse u no e ," we e needed o an accu acy a e o 64.1% .
Simila ly, Gadžo [5] achie ed a high accu acy a e o 98-100% using eigh pa ial indica o s
om he Beneish M-sco e model. So munen [28] u he no ed ha he classi ica ion abili y o
ce ain inancial a ios may diminish o e ime, sugges ing he need o ongoing e alua ion. In
ou li e a u e e iew, we did no iden i y s udies ha p edic inaccu a e inancial s a emen s
based on ac ual alues o balance shee i ems a he beginning and end o he pe iod (wi hou
using a io indica o s). Tha is he esea ch gap ha his scien i ic pape aims o ill. The basis
o de e mining inaccu a e inancial s a emen s in go e nmen -owned en e p ises is he analysis
o audi o s’ indings and c i icisms, as well as he w i en g ounds o issuing opinions on he i-
nancial s a emen s. Acco ding o ou own esea ch [5], he mos common causes o i egula i ies
in inancial epo ing o go e nmen -owned en e p ises in FBiH include inadequa e accoun ing
es ima es o accoun s ecei able om cus ome s, inadequa e alua ion o ixed asse s, in en-
o y, and p o isions. A ange o me hodologies ha e been p oposed o p edic ing he audi o ’s
opinion on inancial s a emen s. Sánchez-Se ano e al. [27] and S anišić, Radoje ić & S anić
[30] bo h highligh he use o a i icial neu al ne wo ks and machine lea ning algo i hms. Da a
mining can be an ex emely use ul ool o de ec ing i egula i ies wi hin a la ge olume o da a
in inancial s a emen s. La ge amoun s o da a o en con ain hidden pa e ns and ends ha
may indica e po en ial audulen ac i i ies. The e o e, da a mining is used o ex ac knowl-
edge om as da a se s in o de o iden i y beha io al pa e ns ha may indica e aud. The
s udy conduc ed by e alua ed he capabili y o a ious Da a Mining classi ica ion me hods o
iden i y companies ha eleased audulen inancial s a emen s (FFS), wi h a pa icula em-
phasis on ecognizing he key ac o s linked o such aud. The s udy explo ed he applica ion
o Decision T ees, A i icial Neu al Ne wo ks (ANN), and Bayesian Belie Ne wo ks as ools
o de ec ing audulen inancial epo ing. The esul s demons a ed ha he Bayesian Belie
Ne wo k model achie ed he bes pe o mance, success ully classi ying 90.3% o he alida ion
sample in a 10- old c oss- alida ion p ocedu e. Acco ding o he esea ch conduc ed by [26],
u ilized da a mining echniques, including Mul ilaye Feed Fo wa d Neu al Ne wo k (MLFF),
Suppo Vec o Machines (SVM), Gene ic P og amming (GP), G oup Me hod o Da a Han-
dling (GMDH), Logis ic Reg ession (LR), and P obabilis ic Neu al Ne wo k (PNN), o iden i y
3
CRORR 16:1 (2025), 1–15 Gadžo e al.: Da a mining app oach in de ec ing inaccu a e inancial s a emen s...
companies in ol ed in inancial s a emen aud. These echniques we e es ed on a da ase
o 202 Chinese companies and compa ed wi h and wi hou ea u e selec ion. Wi hou ea u e
selec ion, PNN ou pe o med all o he echniques, while wi h ea u e selec ion, bo h GP and
PNN achie ed nea ly equal accu acy, ou pe o ming he o he s. Acco ding o he esea ch con-
duc ed by [21] on inancial s a emen aud, u ilizing da a mining me hods including logis ic
eg ession, decision ees (CART- C4.5. algo i am), and a i icial neu al ne wo ks (ANN), he
esul s indica e ha a i icial neu al ne wo ks and decision ees esul ed in much mo e accu a e
classi ica ion compa ed o logis ic eg ession. Acco ding o he esea ch conduc ed by [29], he
me hods used in he aud de ec ion p ocess we e Linea Reg ession, A i icial Neu al Ne wo ks
(ANN), k-Nea es Neighbo s (KNN), Suppo Vec o Machines (SVM), Decision S ump, M5P
T ee, Random Fo es , and J48. The esul s om he expe imen s indica ed ha da a mining
me hods we e able o de ec he aud ac o s be ween he inancial s a emen s and he e-ledge .
In his s udy, he Decision S ump Algo i hm exhibi ed he bes pe o mance. In addi ion o
he men ioned s udies, nume ous o he esea ch s udies ha e ocused on he e ec i eness o
di e en da a mining echniques in de ec ing aud in inancial s a emen s [17]. Ta usch e al.
[31] in oduced a modi ied e sion o DBSCAN, a densi y-based clus e ing algo i hm, which
ou pe o med p io me hods in de ec ing es a ed inancial s a emen s. Ra isanka e al. [26]
and Gill & Gup a [7] bo h ound ha p obabilis ic neu al ne wo k (PNN) and neu al ne -
wo k echniques we e e ec i e in iden i ying companies eso ing o inancial s a emen aud.
These s udies collec i ely sugges ha a combina ion o clus e ing and classi ica ion echniques,
pa icula ly hose ha inco po a e empo al a ia ion and inancial a ios, can be e ec i e in
p edic ing inaccu a e inancial s a emen s. All o hese au ho s ha e in es iga ed aud de-
ec ion in he p o i sec o . Howe e , We we e unable o iden i y signi ican scien i ic s udies
on de ec ing inaccu a e inancial s a emen s in go e nmen -owned en e p ises. Nume ous au-
ho s ha e demons a ed he e ec i eness o u ilizing he Beneish M-Sco e model in p ac ice,
al hough hey did no employ da a mining echniques bu a he elied solely on he ma he-
ma ical o mula o he model [10].
3. Resea ch elabo a ion
3.1. Me hodology
This pape examines he da a ex ac ed om he inancial s a emen s o Go e nmen -Owned
En e p ises (GOEs) in he FBiH and he co esponding audi epo s p epa ed by he Audi
O ice o he Ins i u ions in FBiH. The s udy encompasses a ime span o 16 yea s, om 2004
o 2019. I comp ises a o al o 200 inancial s a emen s and hei associa ed audi epo s,
all o which we e a ailable a he ime o conduc ing he s udy. Financial s a emen s se e
as he undamen al basis and s a ing poin o analyzing business ope a ions and assessing
he condi ion o a company. They can also be ega ded as con iden ial eco ds gene a ed by
o ganiza ions ha con ain hei inancial ansac ions, including expenses, ealized p o i s, in-
come om loans, e c. [8]. Financial s a emen s p o ide an indica ion o he o ganiza ion’s
inancial eali y and also include managemen no es on business pe o mance and p ojec ed
u u e ends. Fu he mo e, inaccu a e inancial s a emen s decei e use s o inancial epo s
by c ea ing he imp ession ha o ganiza ions a e pe o ming a o ably. The ask o audi ing
is o p o ec he in e es s o capi al owne s and p o ide a eliable in o ma ion ounda ion o
a ional decision-making and managemen o s a e-owned companies. In acco dance wi h he
In e na ional S anda d on Audi ing (ISA) 240 (sec ions 2 and 3), miss a emen s in inancial
s a emen s can occu due o ei he aud o e o ( he In e na ional S anda ds o Sup eme
Audi Ins i u ions-ISSAI does no speci ically de ine inaccu a e inancial s a emen s like ISA
240 does). The key di e en ia ing ac o be ween aud and e o lies in whe he he unde ly-
ing ac ion ha leads o he miss a emen in inancial s a emen s is delibe a e o unin en ional.
4
CRORR 16:1 (2025), 1–15 Gadžo e al.: Da a mining app oach in de ec ing inaccu a e inancial s a emen s...
While aud is a b oad legal concep , audi o s, unde he ISAs, ocus on aud ha causes signi -
ican miss a emen s in inancial s a emen s. The e a e wo ypes o in en ional miss a emen s
ha a e ele an o audi o s: miss a emen s a ising om audulen inancial epo ing and
miss a emen s esul ing om misapp op ia ion o asse s. Al hough audi o s may suspec o ,
in a e cases, iden i y ins ances o aud, hey do no make legal de e mina ions ega ding he
occu ence o aud. Consequen ly, he e m "miss a emen " will be used going o wa d wi hou
explici ly speci ying whe he an e o is in en ional o no . Pu suan o In e na ional S anda d
on Audi ing (ISA) 240 (sec ions 2 and 3), miss a emen s in he inancial s a emen s can a ise
om ei he aud o e o (In e na ional S anda ds o Sup eme Audi Ins i u ions-ISSAI does
no speci ically de ine inaccu a e inancial s a emen s like ISA 240 does). The key di e ence
be ween aud and e o lies in he in en behind he ac ion ha leads o inaccu acies in he
inancial s a emen s—whe he i is in en ional o unin en ional. Al hough he concep o aud
is b oad in legal e ms, unde he In e na ional S anda ds on Audi ing, he audi o ocuses
on auds ha cause ma e ial miss a emen s in he inancial s a emen s. The e a e wo ypes
o in en ional miss a emen s ele an o he audi o : hose a ising om audulen inancial
epo ing and hose esul ing om he misapp op ia ion o asse s [11]. Al hough he audi o
may suspec o , in a e cases, iden i y he occu ence o aud, he audi o does no make legal
de e mina ions o whe he aud has ac ually occu ed. The e o e, he e m miss a emen shall
be used onwa ds wi hou speci ying i e o is in en ional o no .
In e na ional S anda d on Audi ing (ISA) 705 (pa ag aph 2) de ines h ee ypes o modi ied
opinions: a quali ied opinion, an ad e se opinion, and a disclaime o opinion. The decision
ega ding which ype o modi ied opinion is app op ia e depends on he na u e o he ma e
causing he modi ica ion, i.e., whe he he inancial s a emen s a e ma e ially miss a ed o , in
cases whe e su icien and app op ia e audi e idence canno be ob ained, whe he hey could
be ma e ially miss a ed; and he audi o ’s judgmen ega ding he pe asi eness o he e ec s
o possible e ec s o he issue on he inancial s a emen s [12]. Ou model uses balance shee
i ems om inancial s a emen s as p edic i e a ibu es and he ype o opinion on he inancial
s a emen s as he a ge a iable. As he inpu se o da a o he model, we used 24 balance
shee posi ions: Opening and closing balances o accoun s ecei able (a ibu e code: A3, A1),
Sales income o he cu en and p e ious accoun ing pe iod (A4, A2), Ope a ing expenses o he
cu en and p e ious accoun ing pe iods (A6, A5), Opening and closing balances o ope a ing
asse s (A11, A7), Opening and closing balances o P ope y, Plan , and Equipmen (A12, A8),
Opening and closing balances o sho - e m inancial in es men s (A13, A9), Opening and
closing balances o business asse s (A14, A10), Opening and closing balances o Dep ecia ion
(A15, A16), Adminis a i e expenses o he cu en and p e ious accoun ing pe iod (A17, A18),
Figu e 1: Audi epo s dis ibu ion o GOEs in FBiH.
5
CRORR 16:1 (2025), 1–15 Gadžo e al.: Da a mining app oach in de ec ing inaccu a e inancial s a emen s...
Opening and closing balances o sho - e m liabili ies (A21, A19), Opening and closing balances
o long- e m liabili ies (A22, A20), Business p o i /loss o he cu en accoun ing pe iod (A23)
and Ne cash low om ope a ing ac i i ies o he cu en accoun ing pe iod (A24). These
balance shee i ems we e selec ed based on he ac ha audi o s ha e p edominan ly iden i ied
i egula i ies in he alua ion o hese balance shee posi ions as g ounds o issuing quali ied
opinions. Audi o s cu en ly highligh non-compliance wi h he p o isions o IFRS 9, IFRS
15, IAS 2, IAS 16, IAS 36, IAS 37, IAS 38, as well as cash lows om ope a ing ac i i ies
and business esul s in he inancial s a emen s o go e nmen -owned en e p ises in FBiH. The
ou pu a iable was he quali y o inancial s a emen s measu ed ela i e o he audi epo o
s a e audi o s. The esul s o he inal audi epo s o GOEs subjec o he analysis a e gi en
in Figu e 1. Numbe o en e p ises is p esen ed on he x-axis.
The ou pu a iable – audi epo o GOEs in FBiH can be g ouped as ollows:
•As ou ca ego ies o classes, so ha audi epo is a class, as p esen ed in Table 1,
•As wo classes coded as YES ca ego y – unquali ied and quali ied opinion, and NO ca ego y
– ad e se opinion and disclaime o opinion, as p esen ed in Table 2.
Class Audi epo Sample numbe Pe cen age
1 Unquali ied opinion 21 10.50%
2 Quali ied opinion 106 53.00%
3 Disclaime o opinion 6 3.00%
4 Ad e se opinion 67 33.50%
SUM 200 100.00%
Table 1: Fou classes, acco ding o he inal audi epo .
Class Assessmen Sample numbe Pe cen age
1 No 73 36.50%
2 Yes 127 63.50%
SUM 200 100.00%
Table 2: Two classes, acco ding o he inal audi epo .
I is e iden ha p edic ion e o in he i s case would be much highe due o di e en
dis ibu ion o he inal audi epo by classes. Hence, his esea ch ga e ad an age o he
second case. The en e p ises a e di ided in o wo basic g oups:
•The i s g oup includes he en e p ises whose inancial s a emen s we e gi en unquali ied
o quali ied opinion (127 en e p ises).
•The second g oup includes he en e p ises whose inancial s a emen s we e gi en ad e se
o disclaime o opinion (73 en e p ises)
Such o mula ion o he ou pu a iable ca ego izes he p oblem as classi ica ion p oblem,
whe e he aim o he model is o lea n how o ecognize he p ope classi ica ion o he inal audi
epo . The p ima y goal o p edic ion is o de elop a model ha de i es insigh s abou a speci ic
cha ac e is ic o he dependen a iable by u ilizing a combina ion o independen a iables.
P edic i e modeling in ol es de e mining he ou pu a iable o a cons ained da ase , whe e
he symbols ep esen he alues o he ou pu a iable in pa icula ins ances. The choice o
a iables om he a ailable da ase signi ican ly in luences he p ecision and accu acy o he
esul ing p edic i e models.
6
CRORR 16:1 (2025), 1–15 Gadžo e al.: Da a mining app oach in de ec ing inaccu a e inancial s a emen s...
3.2. Da a mining
Da a mining, a ield o knowledge disco e y in da abases [1], can be u ilized o disco e ing
inancial auds. Di e en ypes o da a mining echniques can be employed o his pu pose.
Classi ica ion is a me hod used o e alua e and ind a unc ion ha assigns i ems om a
da a se o p ede e mined classes o he ou pu a iable, based on he inpu a iables’ alues
[18]. Th ough classi ica ion, models can be c ea ed o classi y unknown da ase s in o speci ic
ca ego ies o classes [18]. The classi ica ion p ocess ypically in ol es he ollowing s eps:
•Selec ing classi ie s o implemen ing he classi ica ion algo i hm.
•Choosing he class a ibu e (ou pu a iable).
•Di iding da a se s in o wo: aining da a and es da a.
•T aining he classi ie s on he aining da a se wi h known alues o he class a ibu e.
•Tes ing he classi ie s on he es da a se wi h hidden alues o he class a ibu e.
In he classi ica ion p ocess, exis ing echniques a e applied o e alua e he p oposed p edic-
ion model using collec ed ins ances. In he case o selec ing a classi ie o da ase s ob ained
om he inancial s a emen s o go e nmen -owned en e p ises, he si ua ion is qui e clea . This
da ase is e y small; he p ocedu e o conduc ing he expe imen a ion phase is mo e di icul
due o he ac ha he da a is dynamically changeable. Financial s a emen s da a a e mos ly
o a nume ical and ca ego ical ype, and since hey a e manually ex ac ed om da abases o
inancial s a emen s, hey equi e less cleaning in he p ep ocessing phase. Some applica ions o
di e en classi ie s on such da a a e desc ibed in he pape s men ioned in he lis ed e e ences.
In he pape [15], he au ho s employed DT model o inancial aud de ec ion in compa-
nies. Jan [14] used decision ees in combina ion wi h o he da a mining echniques o achie e
a high accu acy a e o 90.83% in de ec ing inancial s a emen s aud. Ki kos, Spa his &
Manolopoulos [15] also explo ed he e ec i eness o decision ees in de ec ing audulen inan-
cial s a emen s, compa ing hei pe o mance wi h o he da a mining echniques. These s udies
collec i ely highligh he po en ial o decision ees in de ec ing aud in inancial s a emen s
o go e nmen -owned en e p ises. Howe e , some ecen app oaches ha e been in oduced o
inancial aud de ec ion [22]. Kı da & Özçelik [16] ound KNN o be a highly e ec i e classi-
ie , achie ing an accu acy a e o 91.73% in de ec ing inancial s a emen aud. Yao e . al.
[36] also highligh ed he e ec i eness o KNN, pa icula ly when combined wi h suppo ec o
machine (SVM) and s epwise eg ession, in de ec ing audulen inancial s a emen s. These
s udies collec i ely unde sco e he po en ial o KNN in de ec ing aud in go e nmen -owned
en e p ises . A ange o s udies ha e demons a ed he e ec i eness o Mul ilaye Pe cep on
(MLP) in de ec ing aud in inancial s a emen s. T iguei os & Sam [32] and Muba ek & Adali
[24] bo h ound ha MLP, when used in conjunc ion wi h o he machine lea ning echniques,
ou pe o med adi ional me hods in aud de ec ion. Ra isanka e al. [26] and Kwon & Fe oz
[19] u he suppo hese indings, wi h Ra isanka [26] no ing he supe io pe o mance o
MLP in iden i ying companies engaged in inancial s a emen aud, and Kwon [19] epo ing
an 88% accu acy a e in p edic ing SEC in es iga ion a ge s using MLP. These s udies collec-
i ely highligh he po en ial o MLP in de ec ing aud in inancial s a emen s, pa icula ly in
go e nmen -owned en e p ises. Fo example, some s udies ha e employed Bayesian ne wo ks
o de ec inaccu a e inancial s a emen s. Deng [4] ound he algo i hm o be e ec i e in his
con ex , no ing i s po en ial o p oac i e aud de ec ion. Handoko, Wiya di & Handoko [9]
also ound success in using he Beneish M-Sco e me hod, which includes a iables ela ed o
inancial s a emen manipula ion, in de ec ing aud in Indonesian go e nmen -owned en e -
p ises. In his wo k, and in acco dance wi h he abo e men ion, he ollowing algo i hms we e
chosen: Decision T ees (J48) [14,15], K-nea es neighbo s [16,36], Neu al Ne wo k (MLP)
7
CRORR 16:1 (2025), 1–15 Gadžo e al.: Da a mining app oach in de ec ing inaccu a e inancial s a emen s...
[8,19,24,26,32], and Bayesian ne wo k (Nai e Bayes) [4,9] o applica ion on he aining
da ase . The ollowing subsec ion o e s an de ailed o e iew o he machine lea ning echniques
applied in his esea ch o de ec inancial audulen ac i i ies. Decision T ees (DT) a e one
o he mos well-known classi ica ion echniques o en used o model da a in he o m o a ee
s uc u e. These algo i hms es ablish ela ionships be ween inpu ea u es and ou pu s using
a ee-like s uc u e, named o i s esemblance o an in e ed ee. The name "decision ee"
de i es om he ac ha i esembles an in e ed ee. The mos commonly used and widely
ecognized decision ee algo i hm is C4.5, and i s implemen a ion in he Weka so wa e ool is
known as he J48 algo i hm. The J48 (C4.5) algo i hm [25] p esen s an ex ension o P o esso
Ross Quinlan’s ea lie ID3 algo i hm, and is known o i s excep ionally high accu acy. The
ad an age o he J48 algo i hm is he abili y o wo k wi h nume ical and ca ego ized da a [34],
in addi ion, i is easy o implemen and e ec i ely deals wi h noise and missing alues [23],
and i has he abili y o display esul s g aphically. Basic cons uc ion o J48 algo i hm uses a
me hod known as di ide and conque o display ou pu [34,23]:
•Selec he da ase as inpu o he ule-making p ocess.
•Calcula e he no malized in o ma ion gain o each a ibu e.
•Choose he a ibu e wi h he maximum in o ma ion gain as he bes a ibu e. This
a ibu e becomes he oo node and co esponding o he bes p edic o .
•Repea he abo e s ep un il a s opping c i e ion is me , calcula ing he in o ma ion gain
o each a ibu e and adding ha a ibu e as a child node.
Jus as a ee s a s om he oo , b anches in o indi idual b anches, and ends in lea es,
decision ees use b anches o ep esen decision pa hs, wi h he inal ou come ep esen ed by
he lea es. The inal esul is a ee wi h decision nodes and lea es. Each decision node has
wo o mo e di isions, while each lea ep esen s an ou come o decision.
K-nea es neighbo s (KNN) algo i hm ep esen s a s aigh o wa d and easily unde -
s andable supe ised lea ning me hod commonly applied in bo h classi ica ion and eg ession
p oblems . I belongs o a g oup o algo i hms known as ins ance-based lea ning, some imes
e e ed o as ‘lazy lea ning me hods’ because he p ocessing o he aining da a is delayed un il
a es ins ance needs o be classi ied. KNN ope a es on he p inciple ha objec s wi h simila
cha ac e is ics end o belong o he same class o ha e simila ou pu alues. The main goal o
KNN is o g oup n objec s in o k g oups (classes) based on hei a ibu es o ea u es. When
a es ing example is conside ed, i is placed in an n-dimensional me ic space o a ibu e al-
ues. The algo i hm hen de e mines he dis ance be ween he es ing example and all aining
examples in his space. Fo classi ica ion, he mos popula classi ica ion among he k nea es
aining examples is he a ge es ima ion o he classi ie . To de ine ‘nea es ,’ a ious me ics
can be used, including he s anda d Euclidean dis ance and Hamming dis ance, wi h Euclidean
dis ance being commonly used. I k is g ea e han 1, hose close o he es ing example will
ha e a g ea e weigh in he classi ica ion. The p ocedu e s a s wi h an ini ial di ision o
he se i ems in o a selec ed numbe o g oups. The dis ance be ween e e y objec and e e y
g oup is de e mined [22], and objec s a e loca ed in o g oups closes o hem based on gi en
cha ac e is ics. A e joining an objec o a g oup, he cen oid o he g oup is ecalcula ed.
The dis ance o e e y objec om he g oup cen oid is ecalcula ed, and he dis ibu ion o
objec s among g oups con inues un il he selec ed unc ion o he c i e ion sugges s o he wise.
Because o all he abo e, he KNN algo i hm is easy o implemen and unde s and, and is mo e
sui able o small da a se s. Howe e , o la ge da ase s, i can be compu a ionally demanding
because i equi es calcula ing he dis ances o all ins ances in he aining se o each new
ins ance.
8
CRORR 16:1 (2025), 1–15 Gadžo e al.: Da a mining app oach in de ec ing inaccu a e inancial s a emen s...
[33] Wu, H.-P. and Li, L. (2021). The BP neu al ne wo k wi h adam op imize o p edic ing au-
di opinions o lis ed companies. IAENG In e na ional Jou nal o Compu e Science, 48(2). u l:
h ps://www.iaeng.o g/IJCS/issues_ 48/issue_2/IJCS_48_2_16.pd [Accessed 29/6/2024]
[34] Wi en, I. H. and F ank, E. (2005). Da a Mining: P ac ical Machine Lea ning
Tools and Techniques. 2nd Edi ion. San F ancisco: Mo gan Kau mann Publishe s. u l:
h p://academia.dk/BiologiskAn opologi/Epidemiologi/Da aMining/Wi en_and_F ank_Da a
Mining_Weka_2nd_Ed_2005.pd [Accessed 29/6/2024]
[35] Yadia i, W. and Rezwiandha i, A. (2023). De ec ing F audulen Financial Repo ing In S a e-
Owned Company: Hexagon Theo y App oach. JAK (Ju nal Akun ansi) Kajian Ilmiah Akun ansi,
10(1), 128-147. doi: 10.30656/jak. 10i1.5676
[36] Yao, J., Pan, Y., Yang, S., Chen, Y. and Li, Y. (2019). De ec ing audulen inancial s a e-
men s o he sus ainable de elopmen o he socio-economy in China: a mul i-analy ic app oach.
Sus ainabili y, 11(6), 1579. doi: 10.3390/su11061579
[37] Yue, D., Wu, X., Shen, N. and Chu, C.-H. (2009). Logis ic eg ession o de ec ing audulen
inancial s a emen o lis ed companies in China. In: 2009 In e na ional Con e ence on A i icial
In elligence and Compu a ional In elligence. doi: 10.1109/AICI.2009.421
15
