SynDEc: A Synthetic Data Ecosystem

Ka s , Fabian S en; Li, Mahei Manhai; Leimeis e , Jan Ma co
A icle — Published Ve sion
SynDEc: A Syn he ic Da a Ecosys em
Elec onic Ma ke s
P o ided in Coope a ion wi h:
Sp inge Na u e
Sugges ed Ci a ion: Ka s , Fabian S en; Li, Mahei Manhai; Leimeis e , Jan Ma co (2025) : SynDEc: A
Syn he ic Da a Ecosys em, Elec onic Ma ke s, ISSN 1422-8890, Sp inge , Be lin, Heidelbe g, Vol. 35,
Iss. 1,
h ps://doi.o g/10.1007/s12525-024-00746-8
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/323570
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 p://c ea i ecommons.o g/licenses/by/4.0/
Vol.:(0123456789)
Elec onic Ma ke s (2025) 35:7
h ps://doi.o g/10.1007/s12525-024-00746-8
RESEARCH PAPER
SynDEc: ASyn he ic Da a Ecosys em
FabianS enKa s 1 · MaheiManhaiLi1,2· JanMa coLeimeis e 1,2
Recei ed: 22 Ma ch 2024 / Accep ed: 2 Decembe 2024 / Published online: 25 Janua y 2025
© The Au ho (s) 2025
Abs ac
Gi en he c i ical ole o da a a ailabili y o g ow h and inno a ion in inancial se ices, especially small and mid-sized
banks lack he da a olumes equi ed o ully le e age AI ad ancemen s o enhancing aud de ec ion, ope a ional e i-
ciency, and isk managemen . Wi h exis ing solu ions acing challenges in scalabili y, inconsis en s anda ds, and complex
p i acy egula ions, we in oduce a syn he ic da a sha ing ecosys em (SynDEc) using gene a i e AI. Employing design
science esea ch in collabo a ion wi h wo banks, among hem UnionBank o he Philippines, we de eloped and alida ed
a syn he ic da a sha ing ecosys em o inancial ins i u ions. The de i ed design p inciples highligh syn he ic da a se up,
aining con igu a ions, and incen i iza ion. Fu he mo e, ou indings show ha smalle banks bene i mos om SynDEcs
and ou solu ion is iable e en wi h limi ed pa icipa ion. Thus, we ad ance da a ecosys em design knowledge, show i s
iabili y o inancial se ices, and o e p ac ical guidance o p i acy- esilien syn he ic da a sha ing, laying g oundwo k
o u u e applica ions o SynDEcs.
Keywo ds Syn he ic da a· Da a sha ing pla o m· Da a ecosys em· Financial se ices· Da a sca ci y
JEL classi ica ion M15
Mo i a ion
In he wake o ecen global c ises, he enhancemen o
inancial se ices has become a c ucial d i e o accele a -
ing economic eco e y, pa icula ly in de eloping economies
whe e hese se ices a e essen ial o expanding inancial
inclusion and os e ing socioeconomic g ow h (Demi güç-
Kun e al., 2022; Paza basioglu e al., 2020; Whi e e al.,
2021). Howe e , gi en he inancial se ices indus y’s eli-
ance on in o ma ion, inc easing da a a ailabili y is key o
success. This is especially ue o smalle inancial ins i u-
ions, which lack he necessa y olume o high-quali y da a
o le e age cu en AI model ad ancemen s. This lack o
da a esul s in missed oppo uni ies, wi h de eloping coun-
ies po en ially losing ou on up o 5% o GDP h ough
imp o emen s in aud p o ec ion, ope a ional e iciency, and
wo k o ce alloca ion (Whi e e al., 2021; Zacha iadis, 2020).
Al hough he sha ing o inancial ansac ion da a could
educe isks and imp o e anspa ency (B odsky & Oakes,
2017), he eby d i ing economic g ow h (O’Lea y e al.,
2021), i aces signi ican obs acles ela ed o p i acy egu-
la ion and in o ma ion secu i y. Exis ing solu ions such as
open banking and ede a ed lea ning ha e signi ican limi a-
ions. Open banking, which enables cus ome -app o ed da a
exchange be ween inancial ins i u ions, o en p oduces un e-
liable da a due o selec i e pa icipa ion (He e al., 2023) and
lacks co e age o B2B ansac ions (P eziuso e al., 2023).
Fede a ed lea ning, an app oach o aining a model wi h-
ou di ec da a exchange, aces scalabili y issues, es ic s
Responsible Edi o : Ge o S obel
* Fabian S en Ka s
[email p o ec ed]h
Mahei Manhai Li
[email p o ec ed]; [email p o ec ed]
Jan Ma co Leimeis e
[email p o ec ed]h; [email p o ec ed]
1 Uni e si y o S .Gallen, Ins i u e o In o ma ion Sys ems
andDigi al Business, Du ou s asse 50, 9000S .Gallen,
Swi ze land
2 Uni e si y o Kassel, In o ma ion Sys ems, P annkuchs aße
1, 34121Kassel, Ge many
Elec onic Ma ke s (2025) 35:77 Page 2 o 28
pa icipan s o a single sha ed model, and lacks adap abili y
(Baabdullah e al., 2024; Cha e jee e al., 2024). The e o e,
esea ch is equi ed o explo e da a ecosys ems ha acili a e
he exchange o da a be ween inancial ins i u ions and egu-
la o y bodies while sa egua ding he p i acy o indi idual
use s’ in o ma ion (Asse a, 2020).
In he pu sui o es ablishing such an ecosys em enabling
inancial da a sha ing, he applica ion o syn he ic da a gen-
e a ion eme ges as a p omising solu ion. Syn he ic da a,
cu en ly p ima ily used in inancial se ices o ackle class
imbalance in aud de ec ion models by syn hesizing new
audulen samples (Cha i ou e al., 2021), p oduces a i icial
da a ha i done co ec ly main ains p i acy while cap u ing
and gene alizing he pa e ns and a ibu es essen ial o he
aining o machine lea ning models. Combining his wi h
da a sha ing enables he c ea ion o a secu e and obus da a
ecosys em.
While plen y o esea ch on syn he ic da a gene a ion
exis s, signi ican gaps emain o i s p ac ical applica ion
wi hin da a ecosys ems. Resea ch has la gely ocused on
algo i hm de elopmen , lea ing c i ical ques ions unan-
swe ed abou how o design an ecosys em o p i acy-p e-
se ing da a exchange wi h he capabili y o handle complex
da a and achie e in e ope abili y ac oss ins i u ions (Oli ei a
e al., 2019). Addi ionally, he e is limi ed guidance on
which algo i hms a e mos e ec i e in a con ex whe e syn-
he ic da a is le e aged o be sha ed be ween ins i u ions
and no me ely used o inc ease he amoun o aining da a
(Lange in e al., 2022). P ac ical s a egies o in eg a ing
sha ed syn he ic da a wi hin machine lea ning models a e
also spa se, hough such s a egies a e essen ial o ealiz-
ing syn he ic da a’s po en ial in AI applica ions (Sa a o
e al., 2023; S elcenia & P akoonwi , 2023). Finally, incen-
i es, o big as well as small playe s, necessa y o encou -
age pa icipa ion in a syn he ic da a-sha ing ecosys em
emain unde explo ed, despi e being i al o os e ing he
coope a i e engagemen on which such ecosys ems depend
(Gelhaa & O o, 2020). In esponse, ou esea ch seeks o
answe he ollowing ques ions: Wha a chi ec u e is bes
sui ed o secu e da a exchange? Which algo i hms a e mos
e ec i e o da a gene a ion? Wha a e he op imal s a egies
o u ilizing sha ed syn he ic da a wi hin indi idual ins i-
u ions? And do he incen i es wi hin such an ecosys em
e ec i ely encou age pa icipa ion? Fu he mo e, he e is a
need o specialized enginee ing and managemen me hod-
ologies ailo ed o he unique demands o inancial se ices,
whe e s ingen p i acy egula ions and he complex na u e
o ansac ion da a in oduce dis inc challenges (Oli ei a
e al., 2019).
Ou esea ch goal is o p o ide design knowledge o a
syn he ic da a ecosys em ha enables inancial ins i u ions
o sha e inancial ansac ion da a and gene a e u ili y om
doing so. Ou s udy con ibu es o he exis ing li e a u e in
wo signi ican ways. Fi s , i ad ances he ield o da a eco-
sys ems by add essing p i acy challenges and explo ing he
use o da a om mul iple ins i u ions o machine lea ning
(B ée e al., 2024). Second, i o e s p ac ical guidance o
inancial ins i u ions on gene a ing and u ilizing syn he ic
da a, including benchma king di e en algo i hms, se ups,
and aining schemes. Gi en he cu en lack o guidance on
he concep ualiza ion and implemen a ion o such sys ems,
his leads us o he ollowing esea ch ques ion:
RQ: How o design a inancial da a ecosys em (SynDEc)
based on syn he ic da a sha ing?
To add ess he RQ, he pape adop s a mul i ace ed
app oach o in es iga e a chi ec u al design decisions. I
encompasses an examina ion o syn he ic da a gene a ion
echniques wi hin he ecosys em, explo es i s implica ions
o aining p edic i e models, and seeks o iden i y and mi -
iga e po en ial challenges o he ecosys em’s s abili y and
unc ionali y. Addi ionally, i assesses he gene alizabili y
o he de i ed p inciples beyond he domain o inancial
aud de ec ion.
The pape is o ganized as ollows: In he nex sec ion,
we p esen an o e iew o da a ecosys ems in inancial se -
ices and syn he ic da a gene a ion. Nex , we ou line, he
Design Science Resea ch Me hodology by Pe e s e al.
(2007), combining con ex -d i en inno a ion and i e a i e
de elopmen , which we use as ou me hodological ounda-
ion. In he i s o ou ou design cycles, we diagnose he
p oblem space h ough he me a (MR) and design equi e-
men s (DR) based on bo h li e a u e and expe in e iews.
Based on his, ou ini ial se o design p inciples (DP) is
de i ed and ins an ia ed as a sys em a chi ec u e. Building
on his he second design cycle e alua es he easibili y o
di e en syn he ic da a gene a ion and in eg a ion me hods.
The ollowing design cycle ex ends his by e alua ing he
p oposed app oach in new domains while also in es iga ing
imp o emen s o he ecosys em based on da a gene a ion and
exchange. Las ly, design cycle ou akes a ne wo k iew,
in es iga ing design elemen s o ensu e ea ly challenges e-
quen ly seen in da a ecosys ems can be o e come. Finally,
we discuss he indings, ou line limi a ions, p o ide a pe -
spec i e o u u e wo k, and conclude wi h a b ie summa y.
Rela ed wo k
Da a ecosys ems in inancial se ices
The g owing ecogni ion o da a as a c i ical asse o inno-
a ion, g ow h, and alue c ea ion has led i ms o inc eas-
ingly seek ex e nal sou ces o enhance hei da a capabili ies
(Bagad e al., 2021; Gelhaa & O o, 2020). One p omising
Elec onic Ma ke s (2025) 35:7 Page 3 o 28 7
app oach is he o ma ion o in e -o ganiza ional ne wo ks,
whe e o ganiza ions collabo a e o sha e esou ces and
knowledge (G ay & Si es, 2013). Wi hin his con ex , da a
ecosys ems ha e eme ged as an e ec i e amewo k o da a
exchange (Abbas e al., 2021; Heinz e al., 2022; Zuide wijk
e al., 2014). De ined as “a se o ne wo ks composed o
au onomous ac o s ha di ec ly o indi ec ly consume, p o-
duce, o p o ide da a and o he ela ed esou ces” (Oli ei a
& Lóscio, 2018, p. 4), da a ecosys ems a e buil a ound ou
key cons uc s: (1) ac o s, (2) hei oles, (3) ela ionships
among hem, and (4) he esou ces hey equi e. Ac o s in
hese ecosys ems—whe he o ganiza ions, indi iduals, o
ins i u ions— ake on oles such as da a consume s, p o id-
e s, and in e media ies, each con ibu ing uniquely o he
ecosys em's unc ion (Oli ei a & Lóscio, 2018; an Schalk-
wyk e al., 2016). The oles hey assume d i e speci ic asks,
such as da a in e media ies connec ing a ious ac o s and
da a consume s analyzing and p o iding eedback o da a
p o ide s. These in e ac ions, and he dependencies ha
a ise om hem, o m he ela ionships ha unde pin he
ecosys em (Heims äd e al., 2014; Oli ei a & Lóscio, 2018).
A he co e o a da a ecosys em, da a pla o ms p o ide he
echnical in as uc u e o p ocessing and managing da a
om di e se sou ces, enabling a ious da a applica ions.
These pla o ms o en inco po a e da a ma ke places, which
se e as sel -se ice pla o ms ha connec da a p oduce s
and consume s (G öge , 2021). Ano he closely ela ed con-
cep is da a spaces, which a e equen ly used o desc ibe
da a-sha ing ecosys ems ac oss o ganiza ions and hus will
be used as synonyms in his pape (O o e al., 2019).
Building on his ounda ion, ecen esea ch has shi ed
i s ocus o he go e nance and ope a ionaliza ion o da a
ecosys ems, pa icula ly in he a eas o da a so e eign y
(Ja ke, 2017) and us (Gelhaa & O o, 2020; Schä e e al.,
2023), which a e c i ical o ensu ing secu e and eliable
da a exchange. Howe e , in hei comp ehensi e e iew o
da a ecosys ems, B ée e al. (2024) iden i ied se e al gaps
wi hin he li e a u e ha a e cu en ly unde - esea ched,
among hem da a secu i y and he in eg a ion o a i icial
in elligence and machine lea ning wi hin da a ecosys ems.
On he one hand, da a secu i y deals wi h ways da a can
be s o ed and sha ed wi hin da a ecosys ems while emain-
ing p o ec ed as well as he in luence o such measu es on
he u ili y o da a ecosys ems (B ée e al., 2024). On he
o he hand, machine lea ning and a i icial in elligence ha e
become cen al o he o ma ion o da a ecosys ems, ye he e
is a need o a deepe unde s anding o he equi emen s o
sha ing AI aining da a and how aining on sha ed da a
should be conduc ed (B ée e al., 2024). Ou esea ch seeks
o add ess hese challenges by p oposing a new ype o da a
ecosys em cen e ed on syn he ic da a, which o e s a means
o mi iga e p i acy isks while main aining he bene i s
o da a sha ing. Addi ionally, we in es iga e s a egies o
maximizing he u ili y o sha ed da a o enhance indi idual
o ganiza ional pe o mance, he eby con ibu ing o bo h he
heo e ical and p ac ical de elopmen o da a ecosys ems.
Wi h cu en esea ch on da a ecosys ems, p edominan ly
concen a ing on applica ions wi hin heal hca e, Indus y
4.0, and sma ci ies (Cappiello e al., 2020), his s udy
ies o ex end his ocus o he inancial se ices indus y.
Gi en he sec o ’s signi ican dependence on highly sensi i e
da a and i s ad anced applica ion o machine lea ning ech-
nologies, his con ex p o ides a sui able se ing o add ess
p e iously iden i ied esea ch gaps in da a secu i y and he
implemen a ion o AI models wi hin da a ecosys ems. Cu -
en esea ch on da a ecosys ems wi hin he inancial se -
ices indus y can be b oadly ca ego ized in o wo esea ch
s eams. The i s s eam cen e s on open banking, a cus-
ome - ocused ecosys em whe e es ablished s anda ds acili-
a e he secu e sha ing o banking da a wi h a ious ac o s
wi hin he inancial se ices ecosys em, based on cus ome
eques s (Cosma e al., 2023). While his app oach g an s
consume s g ea e con ol o e hei da a, i also aises
signi ican da a secu i y conce ns due o he decen alized
na u e o da a s o age ac oss mul iple p o ide s—a c i ical
issue gi en he heigh ened sensi i i y o inancial ansac ion
da a (Y. Wang e al., 2018). Fu he mo e, open banking does
no p o ide ins i u ions wi h an e icien and secu e mecha-
nism o la ge-scale da a exchange, which is essen ial o
applica ions such as aud de ec ion and an i-money launde -
ing (As ow, 2021). The second s eam o esea ch e ol es
a ound ede a ed lea ning, a me hodology ha comple ely
elimina es da a sha ing by enabling dis ibu ed aining o
sha ed models, he eby ensu ing compliance wi h p i acy
p o ec ion egula ions (Awosika e al., 2024; Lei e al., 2023;
Pe ez e al., 2023). Howe e , ede a ed lea ning p esen s
signi ican challenges, including compu a ional o e head,
scalabili y issues, and s ill p i acy isks, as malicious ac o s
migh be able o in e sensi i e da a om he model pa am-
e e s sha ed du ing he aining p ocess (Baabdullah e al.,
2024; Cha e jee e al., 2024). Addi ionally, he necessi y
o pa icipan s in a ede a ed lea ning ecosys em o ag ee
on a single model a chi ec u e, which is di icul o modi y
once es ablished, u he complica es i s implemen a ion.
The cons ain s o exis ing solu ions, coupled wi h he ac
ha da a ecosys ems do no eme ge o ganically bu ins ead
necessi a e s a egic planning a ound a sha ed alue p oposi-
ion, ha e esul ed in he lack o a comp ehensi e inancial
da a ecosys em o da e (Adne , 2017; Immonen e al., 2014).
This is agg a a ed by a esea ch gap in he de elopmen
o specialized enginee ing and managemen me hodologies
ailo ed o he needs o such an ecosys em (Oli ei a e al.,
2019) which a e especially c i ical in he inancial se ices
sec o , whe e s ingen p i acy equi emen s and he com-
plex na u e o inancial ansac ion da a in oduce dis inc
challenges. Consequen ly, u he esea ch is essen ial o
Elec onic Ma ke s (2025) 35:77 Page 4 o 28
add ess hese challenges and o delinea e he a chi ec u al
amewo ks necessa y o he c ea ion o obus and secu e
da a ecosys ems wi hin he inancial indus y.
Syn he ic da a gene a ion andi s applica ion
Syn he ic da a can be de ined as “da a ha has been gene -
a ed using a pu pose-buil ma hema ical model o algo i hm,
wi h he aim o sol ing a (se o ) da a science ask(s)” (Jo -
don e al., 2022, p. 5). This gene a ion p ocess can ake many
o ms as comp ehensi ely ca ego ized by Baue e al. (2024)
in o 20 dis inc me hod ypes. Among hese, gene a i e
ad e sa ial ne wo ks (GANs) a e he mos popula . GANs
lea n by pi ing a gene a o (syn hesizes da a om andom
noise) and a disc imina o (classi ies samples as eal o ake)
agains each o he , esul ing in wo highly skilled ne wo ks
(Good ellow e al., 2014). This a chi ec u e is highly adap -
able, as disc imina o and gene a o can be easily adjus ed o
new asks (e.g., ime se ies o g aph gene a ion) while being
equen ly he bes -pe o ming syn he ic da a gene a ion
me hod (Baue e al., 2024). Ano he commonly employed
syn he ic da a gene a ion me hod is au oencode -based
a chi ec u es, especially a ia ional au oencode (VAE)
(Kingma & Welling, 2013). VAEs a e ained by mapping
an inpu sample o a hidden ep esen a ion, which is hen
mapped back o he o iginal ec o , hus c ea ing a model
ha syn hesizes alid da a om a lowe dimensional ep-
esen a ion. This decode model is hen used o gene a e
da a om andom noise which makes i especially use ul o
lea ning om da a wi h disen angled ea u es (Baue e al.,
2024). Thi d, ecu en neu al ne wo ks, eed o wa d neu al
ne wo ks which include ecu en edges, a e able o gene a e
sequen ial da a o a bi a y leng h. This makes hem ideal o
sequence gene a ion asks such as speech syn hesis, music,
and ime se ies gene a ion (Lip on e al., 2015). Finally, i -
ual en i onmen s a e compu e simula ions in which algo-
i hms in e ac wi h each o he based on p ede ined ules,
gene a ing syn he ic da a in he p ocess (Bonabeau, 2002).
In he con ex o machine lea ning, syn he ic da a is p i-
ma ily u ilized in h ee key a eas: (i) p i a e da a elease, (ii)
da a de-biasing and ai ness, and (iii) da a augmen a ion o
obus ness (Jo don e al., 2022). As he ocus o his pape
is employing syn he ic da a o p i a e da a elease, i will
be in es iga ed in mo e de ail. He eby, p i a e da a elease
desc ibes he case whe e syn he ic da a is used o mi iga e
disclosu e isk, allowing p i acy conce ns and egula o y
issues o be ci cum en ed by subs i u ing eal da a wi h syn-
he ic da a (Es eban e al., 2017; Jo don e al., 2018). How-
e e , his comes wi h ce ain isks o disclosu e, which use s
need o be awa e o . While mul iple isks exis , he mos
ele an is membe ship in e ence which seeks o de e mine
i an indi idual was pa o he o iginal da ase (Bun e al.,
2021; Jo don e al., 2022). This isk is pa icula ly c i ical in
he con ex o inancial ansac ion da a, as e ealing a use ’s
membe ship in a speci ic bank’s da ase could enable mali-
cious ac o s o ca y ou mo e a ge ed audulen ac i i ies,
making aud p e en ion mo e di icul . Resea ch on dealing
wi h membe ship in e ence isks in syn he ic da a, p ima ily
d awn om he heal hca e domain, can be di ided in o wo
majo s eams. The i s s eam ocuses on achie ing gua an-
eed p i acy by modi ying models o con o m o di e en ial
p i acy p inciples, ensu ing bo h he da a and he model a e
p o ec ed. Algo i hms implemen ing his a e he PATE-GAN
(Jo don e al., 2018) o DP2-VAE (Jiang e al., 2022) a chi-
ec u es. The second esea ch s eam ocuses on e alua ing
and managing p i acy isks wi hin accep able limi s o a
gi en olume o published syn he ic da a, p o iding a ious
me ics and h esholds o guidance (H. Chen e al., 2023;
Yan e al., 2022). Popula measu es a e he nea es neighbo
ad e sa ial accu acy isk (Yale e al., 2020), he membe -
ship in e ence isk (Choi e al., 2018), and he meaning ul
iden i y disclosu e isk (Emam e al., 2020). Fu he mo e,
hese measu es ha e also been adop ed by egula o s such as
he Eu opean Medicines Agency and Heal h Canada which
bo h p o ide h esholds o iden i ying disclosu e isk (Yan
e al., 2022).
As he complexi y o models con inues o g ow, neces-
si a ing la ge da ase s, syn he ic da a has been applied in a
a ie y o ields, whe e i is used o acili a e mo e e icien
and e ec i e de elopmen o AI solu ions (Lu e al., 2023).
In inancial se ices, hese ha e been mainly use cases ha
inhibi a s ong class imbalance such as an i-money laun-
de ing and inancial aud de ec ion. He e, syn he ic da a
gene a ion is used o inc ease he amoun o da a wi hin he
mino i y class, he eby inc easing aining e iciency (E. Al -
man e al., 2024; Hilal e al., 2022). The cu en landscape
is la gely domina ed by GAN-based a chi ec u es especially
Wasse s ein GANs due o hei supe io aining s abili y
(Hilal e al., 2022; Se hia e al., 2018; S elcenia & P a-
koonwi , 2023). Howe e , ecen ad ancemen s ha e seen
ans o me -based a chi ec u es (Nicke son e al., 2023) and
di usion-based models (Sa a o e al., 2023) eme ging as
compe i i e al e na i es o GANs. Due o he in e nal usage
o his syn he ic da a, da a p i acy has no been a main con-
side a ion when building hese models. Da a p i acy consid-
e a ions ha e mos ly been explo ed in academic s udies ha
aim o make hei syn he ic da a publicly a ailable. These
s udies ypically employ i ual en i onmen -based sys ems,
such as mul i-agen simula ions, which simula e inancial
ansac ion da a by modeling in e ac ions be ween known
ac o s and beha io s (E. Al man e al., 2024; Jensen e al.,
2023; Lopez-Rojas e al., 2016). While hese app oaches
a e e y secu e om a p i acy pe spec i e as eal da a is
only used du ing model e alua ion o he syn he ic da a,
hey equi e signi ican manual wo k o iden i y pa e ns and
changing beha io s need o be de ec ed i s , be o e hey can

Elec onic Ma ke s (2025) 35:7 Page 5 o 28 7
be in eg a ed in o he simula ion (Baue e al., 2024). How-
e e , he au oma ic gene a ion and sha ing o syn he ic da a
de i ed om eal da a ha e no been ex ensi ely explo ed.
As p i acy conce ns in ensi y due o egula o y p essu e
and cus ome expec a ions, as well as a g owing necessi y
o ex ensi e da ase s o suppo cu ing-edge machine lea n-
ing models (Hi mei e al., 2019), employing syn he ic da a
has he po en ial o add ess p i acy challenges in da a eco-
sys ems. Recen s udies by Sa a o e al. (2023) and Lan-
ge in e al. (2022) ha e begun o in es iga e his po en ial
o inancial se ices. Howe e , hese s udies p ima ily ocus
on compa ing di e en da a gene a ion me hods and p esen
syn he ic da a sha ing as me ely one po en ial applica ion.
This lea es signi ican esea ch gaps ega ding he mecha-
nisms o da a exchange, he op imal s a egies o lea ning
om c oss-ins i u ional syn he ic da a, and he incen i es
o pa icipa ing ins i u ions, eaching beyond inancial se -
ices and ackling cu en challenges in da a ecosys ems in
gene al. Mo eo e , hese s udies o e li le guidance on he
design o such an ecosys em, highligh ing a clea need o
es ablishing design p inciples and bes p ac ices.
Resea ch app oach
A design science esea ch p ojec was ini ia ed o add ess
a esea ch gap in app oaches o enhance p i acy p o ec-
ion wi hin da a ecosys ems while p ese ing da a u ili y
o machine lea ning applica ions. This need, combined
wi h he inancial se ices indus y’s demand o solu ions
o add ess he limi a ions o in e -o ganiza ional collabo-
a ion in ackling inancial aud and an i-money launde -
ing de ec ion, p omp ed he esea ch e o . This p ojec is
aimed a designing an inno a i e a i ac ha p o ides inan-
cial ins i u ions wi h a ool o easily exchange high-quali y
da a wi h each o he enabling hem o inc ease hei aud
and an i-money launde ing de ec ion pe o mance, c ea -
ing guidance on how o implemen such a sys em, as well
as o e alua e i s bene i s and he associa ed p i acy isks
(G ego & He ne , 2013; Pe e s e al., 2007). To achie e
hese objec i es, we adop ed design science esea ch (DSR),
a amewo k pa icula ly sui ed o he i e a i e de elop-
men o no el a i ac s add essing solu ion spaces wi h
b oad implica ions o bo h heo e ical and p ac ical p ob-
lem domains (Pe e s e al., 2007) and p o iding heo e i-
cally jus i ied p esc ip i e knowledge (G ego e al., 2020).
Following his pa adigm, we ocus on c ea ing a i ac s
ha se e o ganiza ional pu pose, in ou case enabling da a
sha ing despi e p i acy es ic ions, h ough a s uc u ed
esea ch p ocess ha igo ously builds and e alua es iable
solu ions (A. R. He ne e al., 2004; Ma ch & Smi h, 1995).
Following Scheide e al. (2023), ou a i ac is a “model”
(Ma ch & Smi h, 1995), a ype o DSR a i ac ha se es
as a simpli ied ep esen a ion o eali y and accumula es
speci ic design knowledge (Ma ch & Smi h, 1995); hus,
DSR p o ides a sui able amewo k o ou s udy (A. R.
He ne , 2007; Ii a i, 2007). Ou model p esen s a s uc-
u ed app oach o designing a da a ecosys em unde p i acy
and da a complexi y cons ain s, exempli ying a solu ion
o he p oblem discussed in he ea lie sec ions. Ou me h-
odological app oach o DSR— he design science esea ch
me hodology (DSRM) by Pe e s e al. (2007) has six s eps,
a anged in sequen ial o de , and inco po a es an i e a i e
esea ch p ocedu e by design. The p ocess ypically s a s
wi h he iden i ica ion o a esea ch p oblem wi h p ac ical
ele ance, in ou case, he challenge o da a sca ci y wi hin
inancial aud de ec ion. Nex , he solu ion objec i es a e
designed o add ess he s a ed challenges and o c ea e a
meaning ul a i ac . In line wi h DSR, he insigh s gained
om he build-and-e alua e p ocess mus be gene alizable
and he e o e applicable in mo e gene ic se ings (Jones &
G ego , 2007). Also, he design a i ac s should esul in
p o ound dis up ions o adi ional ways o doing business
(A. He ne & G ego , 2022). Based on hese objec i es and
on heo y, he a i ac is designed and de eloped in he nex
esea ch p ocess s ep. Phase 5 comp ises e alua ion, which
is necessa y o es whe he an a i ac achie es he pu pose
o i s c ea ion and o p o e his achie emen using igo ous
me hods (Venable e al., 2016). The e alua ion phase also
helps one o be e unde s and he p oblem a hand and hus
o ealize imp o ed ou comes (A. R. He ne e al., 2004).
Due o he i e a i e na u e o his p ocess, i can be epea ed
un il a sui able a i ac is de i ed. The design knowledge in
he o m o DPs wi h hei DRs and MRs gene a ed du ing
his p ocess can be seen as a nascen design heo y, cap-
u ing a gene al solu ion in a class o a i ac s (Baske ille
e al., 2018). While MRs a e high-le el, gene alized goals
ha an a i ac mus sa is y o add ess a class o p oblems,
p o iding he ounda ional objec i es o a i ac design
(Walls e al., 1992), DRs a e speci ic, ac ionable speci ica-
ions ha de ail he necessa y ea u es and cha ac e is ics an
a i ac mus ha e o ul ill he me a- equi emen s (G ego
& He ne , 2013). Las ly, DPs a e p esc ip i e, ac ionable
guidelines de i ed om design equi emen s and g ounded
in bo h heo e ical ounda ions and empi ical e idence, p o-
iding clea ins uc ions o c ea ing a i ac s ha mee he
speci ied equi emen s and add ess he unde lying p oblem
space (G ego e al., 2020). Thus, especially he DPs can be
used o guide ac ions in a wide ange o p oblems, in pa -
icula , da a ecosys ems whe e da a wi h a complex s uc u e
needs o be sha ed unde p i acy es ic ions (A. R. He ne
e al., 2004). They con ibu e o he heo e ical ad ance-
men o he in o ma ion sys ems (IS) communi y and p o-
ide aluable guidance o p ac i ione s in designing simila
a i ac s (Baske ille e al., 2018; Sein e al., 2011). Since
he DSR app oach equi es in eg a ion in o an o ganiza ional
Elec onic Ma ke s (2025) 35:77 Page 6 o 28
con ex , he p ojec was conduc ed in collabo a ion wi h
he UnionBank o he Philippines, a apidly g owing digi-
al bank, as well as a Eu opean neo bank wi h a ocus on
wholesale ansac ion banking. Bo h banks apidly scaled
hei digi al ansac ion in as uc u e in ecen yea s and
a e now looking o new ways o ackle ansac ion aud and
money launde ing. While he banks g an ed us deep insigh s
in o he p oblem o limi ed ansac ion da a and p o ided
in aluable eedback h ough all cycles, i was decided ha
p o o yping and e alua ion would be conduc ed on publicly
a ailable da ase s ins ead o eal bank da a o educe isks
and allow as i e a ions o c ea e a solid unde s anding o
po en ial pi alls.
Wi hin his DSRM amewo k, ou i e a i e design
cycles we e conduc ed, hus allowing o con inuous e ine-
men o he a i ac ’s design based on eedback and de i e
insigh s (Mulla key & He ne , 2019; Sein e al., 2011). In
he nex pa ag aph, he ac i i ies in each cycle a e in o-
duced which a e ou lined in he ollowing g aphic (Fig.1).
Fi s , he DSRM p ojec s a s wi h p oblem iden i ica-
ion and mo i a ion, ocusing on s akeholde p oblems and
challenges. This was done by conduc ing a sys ema ic li e a-
u e e iew on da a ecosys ems, syn he ic da a, and inancial
aud de ec ion as well as semi-s uc u ed in e iews wi h
employees a di e en le els a ou pa ne banks, who a e
engaged in da a sha ing ini ia i es, aud de ec ion o da a
analy ics, and machine lea ning p ojec s. Fu he mo e, hese
in e iews we e used o iden i y he objec i es o ou solu-
ion by de i ing DRs and MRs. Nex , we i e a ed he i s
“Design—Demons a e—E alua e” cycle. In he design
phase, we o mula ed he ini ial se o DPs. These p inci-
ples we e hen ansla ed in o a sys em a chi ec u e du ing
he demons a ion phase, speci ying i s ma e ial p ope ies
like algo i hms and in e ac ion laye s. Subsequen ly, an
e alua ion was conduc ed, in ol ing eedback om aca-
demics and indus y expe s h ough ou semi-s uc u ed
in e iews. The ou comes helped e alua e he easibili y o
he ini ial design and led o he e inemen o selec ed DPs
in he second i e a ion. In cycle 2, we conduc ed a li e a u e
e iew iden i ying sui able algo i hms o syn he ic inancial
ansac ion da a gene a ion and based on hem, ins an ia ed
a p o o ype which was subsequen ly e alua ed on a publicly
a ailable eal-wo ld c edi ca d ansac ion da ase o iden-
i y he mos sui able syn he ic da a gene a ion algo i hm,
es ablish he easibili y o he solu ion, and demons a e he
p i acy-p ese ing p ope ies o syn he ic da a. Based on
addi ional expe eedback as well as wo la ge simula ed
inancial ansac ion da a se s, cycles 3 and 4 e ine he exis -
ing DPs and in oduce new ones whe e needed. While cycle
3 explo es he local le el o he ecosys em in mo e de ail,
cycle 4 ocuses on he global le el and coope a i e chal-
lenges wi hin he ecosys em. Th oughou he DSRM cycles,
Fig. 1 S eps and design cycles wi hin ou design science esea ch s udy based on Pe e s e al. (2007)
Elec onic Ma ke s (2025) 35:7 Page 7 o 28 7
we i e a i ely abs ac ed he equi emen s, DPs, and sys em
ea u es. Thus, ou main heo e ical con ibu ions lie in he
abs ac ed a i ac s, pa icula ly he DPs, which a e i s
de i ed in “Design o ini ial DPs” and con inuously e ined
h oughou he pape .
P oblem iden i ica ion andmo i a ion
The diagnosis phase consis s o wo asks: unde s anding he
p oblem and solu ion domain and de ining he ecosys em’s
equi emen s. Fi s , we posi ioned ou DSRM p ojec wi hin
he domain o in e -ins i u ional collabo a ion wi hin inan-
cial se ices. Wi h a majo ocus o such collabo a ion being
inancial aud de ec ion, a i s li e a u e e iew on da a
ecosys ems, syn he ic da a, and inancial aud de ec ion was
conduc ed. Following he me hodology by Webs e and Wa -
son (2002), ou sea ch s ings we e es ablished (Table1)
and he ollowing da abases: ScienceDi ec , EBSCOhos ,
Sp inge Link, IEEE Xplo e, and AISeL, we e que ied o
a icles con aining he p e iously de ined sea ch s ing in
i le, abs ac , o he au ho keywo ds. Fu he mo e, only
pape s w i en in he English language and published wi hin
he pas 5yea s we e included. This ini ial que y esul ed
in a o al o 3794 pape s, which we e hen il e ed based on
a sc eening o i les and abs ac s. While o pape s iden-
i ied by he “F aud De ec ion” que y s ings only pape s
we e included ha deal wi h inancial ansac ion aud and
ei he ocus on p i acy o a mul i-o ganiza ional con ex ,
o pape s selec ed by he “Da a Ecosys em” s ing he only
inclusion c i e ia we e a ocus on da a ecosys ems. A e
adding mo e ele an pape s h ough a o wa d and back-
wa d sea ch a o al o 61 pape s we e selec ed o inclusion
in he li e a u e e iew.
The analysis o he i s pa o ou li e a u e e iew
ocusing on aud de ec ion e ealed ha he limi ed a aila-
bili y o da a is a signi ican challenge, especially o smalle
o ganiza ions (Kula illeke, 2022; P an o e al., 2022). Espe-
cially wi h inc easingly sophis ica ed ad e sa ies (Qiao
e al., 2024) and hus, mo e complex aud de ec ion models,
equen ly buil based on deep lea ning a chi ec u es, mo e
da a is needed o model aining (Au na e al., 2023; Hilal
e al., 2022). This need o inc easing amoun s o aining
da a is u he agg ega ed by he ex eme class imbalance
o da ase s (la ge da ase s a e needed o a su icien num-
be o samples in he mino i y class) as well as he as -
changing na u e o audulen pa e ns (Abdul Salam e al.,
2024; Ryman-Tubb e al., 2018). Tackling his, equen ly,
he p oli e a ion o c oss-ins i u ional da a is p esen ed as a
po en ial solu ion, o inc ease he amoun o a ailable da a
and ain be e and mo e obus models (Kong e al., 2024;
Myalil e al., 2021; Qiao e al., 2024). Howe e , due o he
high sensi i i y o inancial ansac ions and he connec ed
isk o p i acy leakage, his exchange is usually p ohibi ed
by ex e nal egula ion o in e nal guidelines (Bian & Zheng,
2023; P an o e al., 2022; Ryman-Tubb e al., 2018). To o e -
come his p oblem, equen ly ede a ed-lea ning-based
solu ions a e p oposed, allowing he aw da a o emain
local, while a joined model is ained (Kong e al., 2024;
Lei e al., 2023; P an o e al., 2022). While hese app oaches
show some p omise, hey e ain signi ican d awbacks such
as he compu a ional o e head, scalabili y issues, and he
necessi y o ag ee on a single model a chi ec u e, which is
di icul o modi y once es ablished (Baabdullah e al., 2024;
Cha e jee e al., 2024). This leads us o he conclusion ha
he e is a need o a da a ecosys em ha allows inancial
ins i u ions o exchange da a wi h one ano he while s aying
complian wi h laws and in e nal egula ions on da a p i acy
and gi ing hem he eedom o use his da a o ul ill hei
speci ic needs.
De ini ion o solu ion objec i es
Looking o po en ial solu ions, we d ew on he second
pa o ou li e a u e e iew ocusing on da a ecosys ems
p o iding ele an insigh s on how such challenges can be
na iga ed and po en ially o e come in he con ex o inan-
cial da a. Pa icula ly pape s om he heal hca e domain
(H. Chen e al., 2023; Mo ley-Fle che , 2022), in es iga ions
in o he eme gence (Gelhaa & O o, 2020) and o ganiza ion
(Lange & Mukhe jee, 2023) o da a ecosys ems as well as
Table 1 Resul s o sys ema ic li e a u e sea ch
a De ailed il e c i e ia can be ound a h ps:// anony mous. 4open. scien ce/ / Syn h e icD a aEc osys ems- 801C/ Cycle1_ Ini i alDes ignP incip les/
README. MD
ID Sea ch s ing Hi s Fil e : i leaRemo e dupli-
ca es Fil e :
abs ac aFwd and Bwd
sea ch To al
I “Financial” AND “F aud De ec ion” 2471 336 449 30 5 35
“T ansac ion” AND “F aud De ec ion” 990 139
II “Financial” AND “Da a Ecosys em” 164 13 19 18 8 26
“Syn he ic Da a” AND “Da a Ecosys em” 169 6
Elec onic Ma ke s (2025) 35:77 Page 8 o 28
he p econdi ions o da a sha ing (Fassnach e al., 2023),
we e de imen al in de i ing he design equi emen s p e-
sen ed in he ollowing sec ion.
To ex end ou insigh s in o he domain beyond academic
li e a u e nex , nine semi-s uc u ed in e iews wi h employ-
ees a a ious le els a ou p ojec pa ne s, wi h a ocus on
aud de ec ion o da a science, we e conduc ed ( o de ails,
see Table2). Que ying hem o challenges as well as po en-
ial solu ions o ackling da a sca ci y wi hin hei domain.
Based on his, we o mula ed wo me a- equi emen s
(MR) ha any solu ion mus adhe e o. MR1 emphasizes he
ease o da a sha ing be ween inancial ins i u ions, encom-
passing bo h echnical, legal, and collabo a ion aspec s.
The need o echnical ease o use was in o med by insigh s
d awn om he medical ield, whe e challenges ela ed o
ool a ailabili y and a ying da a s anda ds we e iden i ied
as hind ances o da a sha ing ( an Panhuis e al., 2014).
The legal dimension in ecosys em usabili y was mo i a ed
by di e se egula o y equi emen s ac oss ju isdic ions, as
obse ed in exis ing app oaches o sha ing inancial ansac-
ion da a (Blake e al., 2019). Las ly, ease o collabo a ion
was d awn om he ecosys em li e a u e, whe e coope a i e
challenges we e ou lined as a majo hu dle o da a ecosys em
de elopmen (Gelhaa & O o, 2020). MR2 highligh s he
necessi y o inc eased u ili y as a esul o sha ing da a. This
equi emen emana ed om discussions wi h ou pa ne s
ega ding hei goal o es ablishing a da a-sha ing ecosys em
and om he li e a u e desc ibing incen i es o pa icipa ion
in da a ecosys ems (Gelhaa e al., 2021).
Nex , we e ined he MRs in o mo e speci ic DRs, d aw-
ing om li e a u e as well as he knowledge o ou p ojec
pa ne s.1 To incen i ize use s o pa icipa e in da a-sha ing,
se up as well as eoccu ing cos s need o be as low as pos-
sible, which is e lec ed in MR1 and p opaga es in o DR1
and DR2. This is impo an because while a da a s anda d o
inancial ansac ion da a exis s, di e en banks di e ge om
i (Majo & Mangano, 2020), which was also con i med du -
ing ou in e iews (“Di e en da a p o ide s ha e di e en
schemas and ansac ion languages.”—In e iewee 2); hus,
a da a ecosys em needs o be lexible enough o accommo-
da e a ious inpu da a s uc u es (DR1). This is pa icula ly
impo an as da a needs o be egula ly upda ed and he cos
o hese upda es should be as low as possible. Fu he mo e,
da a p i acy s anda ds imposed by egula o s and in e nal
policies mus be upheld (“In e ms o da a sha ing we do no
engage in any hing, because his is he pain wi h inancial
ins i u ions, we a e eally p o ec i e o ou da a”—In e -
iewee 8–1). Ou in e iews e ealed ha in he con ex o
ou pa ne ins i u ions, his means ha all eal da a mus
be p ocessed locally wi hin he inancial ins i u ion (DR2).
F om a da a-cen ic pe spec i e, he pe o mance o machine
lea ning me hods can be enhanced by inc easing he ol-
ume o aining da a a ailable (Sun e al., 2017). Thus, MR2
can be achie ed by enabling he combina ion o da a om
mul iple sou ces h ough he da a ecosys em and making i
accessible as a uni ied da a sou ce (DR3). Gi en he goal o
c ea ing an ecosys em ha is applicable o mul iple asks, he
absence o a dominan algo i hm in many ields (e.g., aud
de ec ion), and he insigh om ou in e iews ha banks
p e e o build and exclusi ely own hei solu ions (“One
model will no be enough, i will be a collec ion o models
which answe di e en ques ions …”—In e iewee 8–1),
he da a ecosys em mus suppo di e se ypes o algo i hms
(DR4). Addi ionally, he imbalanced na u e o aud da a
necessi a es ools on he ecosys em o add ess da a imbal-
ances h ough il e ing, o e sampling, and unde sampling
(DR7), as mos machine lea ning algo i hms pe o m be e
on balanced da ase s (Longadge & Dong e, 2013). As aud
pa e ns change quickly when disco e ed, he imely in eg a-
ion o ecen aud pa e ns in o aud de ec ion algo i hms
is c ucial (Benchaji e al., 2021; Zhu e al., 2021). As his is
u e ly impo an , wo DRs we e dedica ed o achie ing his.
Fi s , ins i u ions should ha e he capabili y o au oma ically
Table 2 O e iew in e iewees o solu ion equi emen s
*In e iewee om UnionBank o he Philippines
ID Job i le Expe ise Yea s o expe ience Leng h o in e iew
In e iewee 1 Chie da a scien is Da a science 10yea s 00:51:10
In e iewee 2 Senio da a scien is Da a science 5yea s 00:37:34
In e iewee 3 Da a scien is Da a science 5yea s 00:36:30
In e iewee 4 Chie inancial o ice F aud de ec ion > 20yea s 00:38:36
In e iewee 5 Senio compliance o ice F aud de ec ion > 20yea s 00:59:08
In e iewee 6 Junio compliance o ice F aud de ec ion 4yea s 00:44:27
In e iewee 7*Head o he AI cen e o excellence Da a science > 20yea s 00:19:25
In e iewee 8*Head o da a science en u es Da a science 10yea s 00:31:03
1 A de ailed mapping om in e iew quo es o DRs can be ound on:
h ps:// gi hub. com/ Fa um an/ Syn h e icD a aEc osys ems/ blob/ mas e /
Cycle1_ Ini i alDes ignP incip les/ README. MD
Elec onic Ma ke s (2025) 35:7 Page 15 o 28 7
an examina ion o he li e a u e ega ding he enhancemen o
machine lea ning pe o mance h ough he inco po a ion o syn-
he ic da a was conduc ed, aiming o de e mine an op imal a io o
eal o syn he ic da a (mix-in pe cen age). While some esea che s
only o e sample he mino i y class using syn he ic da a (Cha i ou
e al., 2021; S elcenia & P akoonwi , 2023), o he s ain models
exclusi ely on syn he ic da a (Sa a o e al., 2023) o combine
eal wi h syn he ic da a (Dahmen & Cook, 2019). Thus, i emains
unclea i he e is an op imal mix-in pe cen age ha indi idual
ins i u ions should inco po a e in o hei design.
To ind he op imal way o gene a e syn he ic da a o ou
ecosys em, his sec ion in es iga es da a gene a ion con igu-
a ions ha u ilize he en i e da ase as well as hose ained
on dis inc da a subse s and u he analyzes he bene i o
di e en p e-p ocessing s eps du ing he syn hesizing p o-
cess. Due o he challenge o da ase imbalance, models
end o be biased owa ds he majo i y class, dec easing he
quali y o da a in he mino i y class; mi iga ing his issue,
o e sampling can be applied du ing he gene a ion p ocess
o enhance gene a o obus ness, albei a he isk o dis o -
ing da ase composi ion ( oo many posi i e samples) (Ki an
& Kuma , 2024). Second, he cons uc ion o dis inc syn-
he ic da a gene a o s o each class has been p oposed as
an al e na i e solu ion. Enabling he gene a o o be e cap-
u e he cha ac e is ics o each indi idual class. Howe e ,
his esul s in he p oblem ha he mino i y class gene a o
is only ained wi h a small da ase , which migh ha m i s
gene alizabili y (Eile sen e al., 2021). To emedy his, Fan
e al. (2022) ha e sugges ed a no el me hodology whe e he
gene a o o he mino i y class is p e- ained using samples
om he majo i y class, hus ci cum en ing he p oblem.
Demons a ion h oughimplemen a ion
o di e en aining andda a‑ using schemes
In his sec ion, we ope a ionalized he de i ed DPs in o a
p o o ype sys em in Py hon using a modi ied e sion o he
syn he ic da a aul lib a y (Pa ki e al., 2016). Building upon
he a chi ec u e om design cycle wo, he local laye was
modi ied o accommoda e o di e en gene a ion schemes
wi h and wi hou o e sampling as well as p e- aining on
he local le el. Fu he mo e, he aining scheme o he p e-
dic ion model was modi ied so ha he sys em was able o
accommoda e aining wi h di e en mix-in pe cen ages.5
E alua ion o di e en aining andda a‑ using
schemes
One challenge in e alua ing he b oade easibili y o ou
syn he ic da a sha ing ecosys em is he lack o publicly
a ailable inancial ansac ion da a (Jensen e al., 2023).
Howe e , mul iple esea che s ha e shown ha simula ed
inancial ansac ions can be sui able o alida ing new
models o e en e alua ing in e en ions (Lange in e al.,
2022; Sa a o e al., 2023). The e o e, in his as well as
he nex cycle, we will use wo da ase s, one o an i-money
launde ing (IBM-AML6) and one o audulen ansac-
ions (IBM-CCF7), which we e gene a ed by using a mul i-
agen -based app oach, simula ing ac o s ha ac acco ding
o p ede ined ules, hus c ea ing a s eam o ansac ions
(E. Al man e al., 2024; E. R. Al man, 2019). The esul ing
da ase s ha e he ad an age o being magni udes la ge in
size (IBM-AML: 31898238/ IBM-CCF: 24386900) han he
da a used in he p e ious cycle (IEEE-CIS: 1097231) and
ha e a ne wo k s uc u e mo e simila o he one in eal da a.
Howe e , due o i s simula ion-based na u e, i migh no
Fig. 6 Upda ed sys em a chi ec-
u e ( e sion 2)
5 The ull implemen a ion o Cycle 3 can be ound he e: h ps://
gi hub. com/ Fa um an/ Syn h e icD a aEc osys ems/ blob/ mas e / Cycle3-
4_ Ecosy s emE alua ion/ README. MD
6 h ps:// www. kaggle. com/ da as e s/ eal m an2019/ ibm- ans ac io ns-
o - an i- money- laund e ing- aml
7 h ps:// www. kaggle. com/ da as e s/ eal m an2019/ c edi - ca d- ans
ac io ns

Elec onic Ma ke s (2025) 35:77 Page 16 o 28
inhibi all cha ac e is ics ound in eal da a. As he selec ed
da ase s do no include a inancial ins i u ion (IBM-CFF)
o he numbe o inancial ins i u ions p esen in he da a
is oo big (IBM-AML, 122333 di e en banks), he da a
was a i icially g ouped. This was done by segmen ing he
da a based on he loca ion o he indi idual (IBM-CCF)/
bank (IBM-AML) connec ed o a ansac ion, c ea ing clus-
e s ha simula e he ansac ional ne wo ks o hypo he i-
cal inancial ins i u ions. As a esul , he IBM-CCF da ase
included ou inancial ins i u ions wi h a ela i ely e en
da a dis ibu ion, while he IBM-AML da ase eme ged wi h
se en banks o which wo banks held o e 75% o he da a.
This con as in da ase composi ion a o ds a unique chance
o explo e he syn he ic da a sha ing ecosys em’s unc ional-
i y unde a b oad a ay o condi ions. Mo eo e , an analysis
o clien dis ibu ion pos -spli o each p o ide highligh ed
signi ican dispa i ies, aligning wi h he an icipa ed di e si y
suspec ed wi hin mul i-ins i u ional da ase s. De ails on he
speci ic dis ibu ions a e ou lined in Table7.
To limi he a iables o his in es iga ion, he syn he ic da a
gene a ion model and he aud p edic ion model we e kep
cons an . Fo he syn he ic da a gene a ion model, he p e i-
ously supe io TVAE-based gene a o 8 wi h hype pa ame e s
uned o he indi idual ins i u ion was used. Simila o Cycle
2, a XGBoos classi ie wi h hype pa ame e selec ion using
h ee old-c oss alida ion was chosen as he p edic ion model
and he pe o mance compa isons we e done using he ROC
AUC sco e on a holdou da ase (30% o he da a).
Be o e, in es iga ing he app oaches modi ying he local
laye , he ans e abili y o syn he ic ansac ion da a sha ing
beyond ansac ion aud de ec ion was e alua ed (Fig.7). To
do his, we compa ed he a e age ROC AUC sco e be ween
he da ase cons uc ed o inancial aud de ec ion (IBM-
CFF) and he one cons uc ed o an i-money launde ing
de ec ion (IBM-AML).
Figu e7 demons a es ha ac oss bo h da ase s, models
ained wi h syn he ic sha ed da a su passed hose ained
wi hou i , enhancing he ROC AUC sco e by 3.6% in he
ansac ion aud da ase (IBM-CCF) and 6.6% in he an i-
money launde ing da ase (IBM-AML). This e ec can be
conside ed subs an ial wi hin his con ex as e en ecen ly
in oduced aud de ec ion algo i hms o en only inc ease he
ROC-AUC sco e by a ew pe cen age poin s (Hashemi e al.,
2023; Lebicho e al., 2021). This pe o mance gain sugges s
ha he da a ecosys em’s e ec i eness ex ends beyond me ely
de ec ing inancial aud bu is also sui able o o he use cases
u ilizing inancial ansac ion da a such as money launde ing
de ec ion. Thus, con i ming he e sa ili y and po en ial o
he syn he ic da a ecosys em in add essing a b oad ange o
da a challenges in inancial se ices. Subsequen ly, we explo e
whe he a speci ic mix-in pe cen age o eal and syn he ic da a
yields op imal esul s o machine lea ning pe o mance. To
accomplish his, we sys ema ically assess he impac on model
pe o mance by a ying he p opo ion o eal and syn he ic
da a used in aining he models, explo ing a spec um om
0% (no syn he ic da a) o 300% (3 imes as much syn he ic as
eal da a). Figu e8 isualizes his expe imen .
Obse ing he modes upwa d ajec o y o he agg e-
ga ed pe o mance line (black), we can conclude ha he e
is a posi i e e ec o adding syn he ic da a. Howe e , in con-
as o he mo e ola ile pe o mance ends o indi idual
banks (g ey), i appea s he e is no a uni e sally op imal
mix-in pe cen age. Ins ead, dis inc peaks in pe o mance
sugges ha he mos e ec i e mix-in a ios a y by bank.
Consequen ly, we in e ha allowing banks o adjus he
mix-in pe cen age independen ly is mos bene icial. This
insigh has been in eg a ed in o DP5, which manda es ha
banks ha e he au onomy o de e mine hei mix-in a ios,
leading o he upda ed p inciple: DP5—P o ides he capa-
bili y o combine syn he ic da a o ind op imal composi ion
o he aining o machine lea ning models gi en scena ios
wi h da a om mul iple ins i u ions.
Finally, we explo ed a ious con igu a ions and p ep oc-
essing me hods o syn he ic da a gene a ion o o e op imal
guidance o se ing up hese p ocesses a he local le el.
Essen ially, he e a e wo p ima y se ups. The i s , e e ed
o as “ ull,” in ol es aining he syn he ic da a gene a ion
model on he en i e da ase . To mi iga e he isk o he model
p edominan ly gene a ing samples om he majo i y class,
e sions ha andomly o e sample he mino i y class o a
speci ied pe cen age o he da a (“_OS{X}”) while aining
he syn he ic da a gene a o ha e been implemen ed. The
second se up, “sep” en ails aining dis inc gene a ion mod-
els o each class. An ex ension o his app oach, “sepP e”
u ilizes sepa a e gene a o s o each class bu p e- ains he
mino i y class gene a o wi h majo i y class da a. The ou -
comes o hese a ied app oaches a e de ailed in Table8.
The analysis o he da a p esen ed in Table8 yields se -
e al key indings. Ini ially, he “ ull” model demons a es i s
abili y o su pass he baseline pe o mance, ye models buil
on he same aining scheme bu u ilizing o e sampled da a
Table 7 Dis ibu ion o da a ac oss he di e en banks
IBM-CCF IBM-AML
Bank Pc o da a Bank Pc o da a Bank Pc o da a
0 21.54% 0 5.93% 4 29.94%
1 18.58% 1 11.90% 5 45.35%
2 39.16% 2 2.87% 6 2.12%
3 20.72% 3 1.90%
8 A de ailed desc ip ion o he hype pa ame e uning p ocedu e can
be ound he e: h ps:// gi hub. com/ Fa um an/ Syn h e icD a aEc osys
ems/ blob/ mas e / Cycle3- 4_ Ecosy s emE alua ion/ 02_ pa am Sea ch/
README. MD
Elec onic Ma ke s (2025) 35:7 Page 17 o 28 7
exhibi a no able decline in pe o mance. Thus, leading us o
he conclusion, ha o inancial ansac ion da a, o e sam-
pling he da a be o e aining he syn he ic da a gene a ion
model is no sui able. Mo eo e , he “ ull” se up ou pe o ms
con igu a ions whe e syn he ic da a gene a o s a e ained
sepa a ely o each class (“sep”). This subpa pe o mance
s ems om he “sep” model’s poo -quali y syn he ic da a
o he mino i y class, which ails o cap u e aining da a
pa e ns due o limi ed aining da ase size. Howe e , when
he mino i y class model is p e- ained using da a om
he majo i y class (“sepP e”), a signi ican pe o mance
imp o emen is obse ed, su passing all o he me hods.
This enhancemen is p ima ily due o he model’s capaci y
o gene a e highe -quali y samples o he mino i y class wi h
g ea e a iabili y. Fu he discussions wi h pa ne ins i u-
ion expe s emphasized he ad an age o c ea ing class
da a sepa a ely as i enhances p i acy by p e en ing leaks
o sensi i e in o ma ion like aud a es by independen ly
Fig. 7 Compa ison be ween
models ained wi h and wi hou
syn he ic da a o bo h da ase s
Fig. 8 E ec o syn he ic da a mix-in pe cen age on pe o mance
Table 8 Compa ison be ween
di e en syn he ic da a
gene a ion models
Da ase Me hod ROC AUC sco e Da ase Me hod ROC AUC sco e
IBM-AML Wi hou sha ed da a 0.7168 IBM-CCF Wi hou sha ed da a 0.6817
Full 0.7371 Full 0.7042
ullOS_10 0.6435 ullOS_10 0.6618
ullOS_20 0.6199 ullOS_20 0.6360
sep 0.7209 sep 0.6817
sepP e 0.7473 sepP e 0.7323
Elec onic Ma ke s (2025) 35:77 Page 18 o 28
p oducing he samples o each class. Consequen ly, we ha e
e ined DP4 o encapsula e hese insigh s: DP2—P o ide he
sys em wi h he abili y o iden i y, alida e, and apply con-
ex -speci ic syn he ic da a gene a ion echniques wi h mu u-
ally ag eed on o e -sampling in o de o emo e p i a e da a,
gi en guidelines o egula ions on da a sha ing.
Cycle 4: Ne wo k e ec s o  inancial da a
sha ing
In cycle ou o ou DSRM p ojec , we del e in o he global
da a laye , guided by he li e a u e and expe insigh s o
add ess coope a i e challenges wi hin he p oposed syn-
he ic inancial da a ecosys ems. Aiming o e ine ou DPs
o enhance he ecosys em’s capabili y o e ec i ely manage
hese challenges.
Design o mechanisms a  heglobal da a le el
Add essing he second aspec o expe eedback and
in o med by he li e a u e on da a ecosys ems, his cycle
ocuses on he global da a laye and i s DPs o ensu e ha
he c ea ed ecosys em is able o handle he challenges o
da a ecosys ems desc ibed by Gelhaa and O o (2020).
Because coope a i e challenges play a dominan ole in
he ea ly s age o an ecosys em, he ollowing cycle will
ocus on hese (Au io & Thomas, 2014). In hei pape ,
Gelhaa and O o (2020) desc ibe ou majo coope a i e
challenges ha need o be add essed o a da a ecosys em
o eme ge success ully. Fi s , i is necessa y o build us
be ween he pa icipan s. Second, i needs o be shown
ha all ac o s bene i om pa icipa ing in he ecosys-
em. Thi d, i is impo an o iden i y he igh numbe o
pa icipan s. Fou h, in e ope abili y needs o be enabled
h ough he ag eemen on s anda ds. Thus, he ocus o
his sec ion is o e alua e exis ing DPs h ough his lens
and analyze whe he e inemen s o addi ional p inciples
a e necessa y o he de elopmen o an ecosys em capa-
ble o e ec i ely add essing hese challenges. Fi s , us
be ween ecosys em pa ne s can be buil in mul iple ways.
On he one hand, us can be inc eased by adequa e con-
ol mechanisms (Geisle e al., 2021), which is al eady
e lec ed in DP3—P o ide he sys em wi h a back- es ing
mechanism in o de o ensu e newly gene a ed syn he ic
da a ma ches in composi ion and aud de ec ion aining
pe o mance wi h eal da a gi en ha da a quali y can-
no be independen ly e i ied which ensu es su icien da a
quali y in he syn he ic da a ecosys em. On he o he hand,
Maja a e al. (2016) show ha in e media ies play a signi i-
can ole in inc easing pa icipan s’ us in an ecosys em.
In he inancial se ices ecosys em, his ole is ypically
held by public egula o s. To incen i ize hem o pa ici-
pa e in he ecosys em and allow hem o ensu e da a quali y
and hus inc ease us , we p opose DP6—P o ide access
o ex e nal collabo a o s, such as egula o s, o le e age
he syn he ic da a wi hin he ecosys em gi en a di e se se
o syn he ic da a a ailable. This gi es egula o s access
o he ecosys em while adhe ing o he exis ing p i acy
measu es. Howe e , i emains unclea i access o pu ely
syn he ic da a can p o ide enough alue and hus incen i -
ize hei pa icipa ion in he ecosys em. The nex challenge
da a ecosys ems ace is ha all ac o s need o bene i om
pa icipa ing in he ecosys em. While we al eady demon-
s a ed in p e ious i e a ions ha ou da a ecosys em is able
o inc ease he o e all pe o mance, i emains unclea how
his pe o mance gain is dis ibu ed be ween ins i u ions.
To add ess his, u he in es iga ion is needed o check i
adjus men s o ou design need o be made o c ea e su i-
cien incen i es o all ins i u ions. Connec ed o his p ob-
lem is iden i ying he igh numbe o pa icipan s. While
ou p e ious cycles show ha he ecosys em is bene icial
i all ins i u ions pa icipa e, i emains unclea i a simila
e ec exis s, i only pa o he ins i u ions is included in
he ecosys em. To inco po a e his in o ou DPs, DP3 was
ex ended o no only desc ibe he moni o ing o ou going
syn he ic da a bu also co e he e alua ion o pe o mance
gained by using he sha ed syn he ic da a om he da a
ecosys em. This esul s in DP3—P o ide he sys em wi h a
back- es ing mechanism in o de o ensu e newly gene a ed
syn he ic da a ma ches in composi ion and aud de ec ion
aining pe o mance wi h eal da a gi en ha da a quali y
canno be independen ly e i ied. The las coope a i e chal-
lenge ha needs o be o e come is in e ope abili y h ough
he ag eemen on s anda ds. A he momen , ha is al eady
inco po a ed in DP1—P o ide he sys em wi h modula
sys ems design in o de o ensu e independence o local
da a and c oss-ins i u ional p oli e a ion o syn he ic da a
gi en ha he aw da a is sensi i e, om a da a pe spec i e
whe e he local laye o he ecosys em is used o align he
da a so ha i can be easily sha ed wi h he sys em la e .
Fu he mo e, we a gue ha c ea ing DPs o he inancial
da a ecosys em con ibu es o he s anda diza ion o he
ecosys em om an in as uc u e and ecosys em pe spec-
i e and hus by c ea ing hese DPs we con ibu e o o e -
coming his challenge.
Demons a ion h ough hein oduc ion
o non‑sha ing en i ies andindi idual pe o mance
benchma king
In his pa , we u he imp o ed he p o o ype de eloped
in Py hon, by al e ing he global da a laye o allow o he
pa icipa ion o en i ies ha do no con ibu e da a. Addi-
ionally, we upda ed he sys em o ack and epo he
Elec onic Ma ke s (2025) 35:7 Page 19 o 28 7
pe o mance o each pa icipa ing ins i u ion, hus allowing
ins i u ions on an indi idual le el o see he pe o mance
gain om engaging in he da a ecosys em.9
E alua ion o ecosys ems wi hnon‑sha ing en i ies
andindi idual pe o mance benchma king
Simila o he p e ious cycle, his e alua ion again u ilizes
he wo syn he ic da ase s (IBM-AML and IBM-CFF), due
o hei high da a quali y, size, and di e si y. Mo eo e , he
ecosys em se up and e alua ion scheme a e adop ed om
he p e ious cycle, u ilizing he “sepP e” aining scheme.
We s a wi h e alua ing DP6, which allows egula o s
o access pu ely syn he ic da a wi hin he da a ecosys em.
To alida e his DP, he syn he ic da a ha is p o ided o
he egula o s need o be o su icien quali y o hem o
de i e meaning ul insigh s and e ec i ely imp o e hei
models. Howe e , as his canno be easily e alua ed, we
use he pe o mance o a p edic ion model ained on he
da a a ailable o he egula o (only syn he ic da a) as
a p oxy o he quali y o he da a. As he a chi ec u e
chosen in Cycle 3 gene a es sepa a e models o di e en
classes, mo e da a o a speci ic class can easily be gene -
a ed. This is especially ele an o cases whe e only syn-
he ic da a is used as no addi ional posi i e samples om
he eal da a exis . Thus, he expe imen conduc ed had
wo s eps. In he i s s ep, egula o models we e ained
on syn he ic da a wi h di e en amoun s o mino i y class
samples (indica ed by OS_{pe cen age o mino i y class
cases}). F om his selec ion, he o e -sampling a io wi h
he bes pe o mance was chosen and compa ed o he
pe o mance o he models ained a he di e en banks,
once ained on a combina ion o eal and syn he ic da a,
and once ained wi h only eal da a. The esul s o his
expe imen can be seen in he ollowing diag am(Fig.9).
The egula o model, ained exclusi ely on syn he ic
da a, exhibi s pe o mance ha , while no ma ching ha
o he bank’s in e nal models ( ained on a mix o eal and
syn he ic da a), emains signi ican . The model app oaches
he pe o mance o he bank’s baseline models ( ained
on eal da a only), as illus a ed in Fig.9. This capabili y
o e s conside able ad an ages o collabo a o s who would
o he wise lack access o such da a. Consequen ly, allowing
egula o s o access syn he ic da a eme ges as an e ec i e
s a egy o os e collabo a ion and enhance us in he
ecosys em. The e o e, DP6—P o ide access o ex e nal
collabo a o s, such as egula o s, o le e age he syn he ic
da a wi hin he ecosys em gi en a di e se se o syn he ic
da a a ailable is alida ed and was added o ou DPs o
syn he ic da a ecosys ems.
Subsequen ly, he adjus men o DP3 is alida ed,
checking i all banks p o i om he syn he ic da a eco-
sys em and e alua ing i he syn he ic ecosys em includ-
ing ewe ins i u ions is s ill able o p o i om he ne -
wo k e ec s o he ecosys em. To in es iga e his, we plo
he pe o mance o each ins i u ion agains i s baseline
(sco e wi hou any a i icial da a), which can be seen
below(Fig.10).
As e idenced in Fig.10 o each single bank in bo h
da ase s, he pe o mance inc eases by combining eal and
syn he ic da a. Fu he mo e, looking a he igh mos panel
o Fig.10, i can clea ly be seen ha he e is a nega i e co -
ela ion (− 0.09) be ween he pe o mance gained by pa -
icipa ing in he ecosys em and he size o he bank. Thus,
showing ha small banks o e p opo ionally p o i om
pa icipa ion, p o iding a clea incen i e o hem o engage
in he ecosys em. Howe e , e en i absolu e pe o mance
gained by bigge banks is lowe , we a gue ha hey s ill
ha e a su icien incen i e o pa icipa e due o hei la ge
olume o ansac ions, whe e e en small changes in he
Fig. 9 Regula o models using di e en esampled da a and pe o mance o he egula o model (only syn he ic da a) s. he bank models
9 The ull implemen a ion o Cycle 4 can be ound he e: h ps://
gi hub. com/ Fa um an/ Syn h e icD a aEc osys ems/ blob/ mas e / Cycle3-
4_ Ecosy s emE alua ion/ README. MD
Elec onic Ma ke s (2025) 35:77 Page 20 o 28
aud de ec ion pe cen age esul in a high absolu e sum
o p e en ed losses. These esul s lead us o he conclu-
sion ha all banks con ibu ing o he ecosys em p o i om
hei in ol emen and hus he designed ecosys em is able o
o e come ano he one o he p e iously ou lined challenges.
Nex , we in es iga e ou syn he ic da a ecosys em o
cases whe e no all ins i u ions engage in syn he ic da a
sha ing. To achie e his, we simula ed en i onmen s, whe e
none, 50%, 75%, o 100% o all banks we e pa o he eco-
sys em. The esul s can be seen in Fig.11.
Despi e he signi ican di e ence be ween he wo da a
se s ega ding hei da a dis ibu ion (wi h IBM-CCF ha ing
an equal dis ibu ion be ween banks, while IBM-AML has a
highly skewed one), we can clea ly see ha in bo h cases, e en
wi h only hal o he banks being pa o he ecosys em (IBM-
CCF: 2 banks/IBM-AML: 3 banks), a signi ican pe o mance
gain is achie ed. Thus, i seems he bene i s o he syn he ic
da a ecosys em can be ealized om an ea ly s age onwa ds,
making i easy o o e come he hu dle o a minimum numbe
o membe s needing o pa icipa e in he ecosys em, hus ack-
ling ano he o he challenges ou lined p e iously.
Summa izing hese esul s, we we e able o demons a e
ha he p oposed da a ecosys em is able o deli e excess
pe o mance o all pa icipan s in he ne wo k on an indi-
idual le el and i can be seen ha e en o da a ecosys-
ems wi h only a ac ion o he ins i u ions pa icipa ing in
syn he ic da a sha ing, s ill a signi ican pe o mance gain
can be achie ed. Fu he mo e, he e seem o be ne wo k
e ec s o some ex en whe e mo e pa ne s in he ecosys-
em inc ease i s o e all u ili y. As hese esul s alida e he
incen i es o pa ne s o pa icipa e in an ecosys em, we
con i m ou DP3—P o ide he sys em wi h a back- es ing
mechanism in o de o ensu e newly gene a ed syn he ic
da a ma ches in composi ion and aud de ec ion aining
pe o mance wi h eal da a gi en ha da a quali y canno
be independen ly e i ied.
Discussion
This esea ch pape is aimed a ex ending he esea ch on
p i acy in da a ecosys ems as well as machine lea ning o
mul i-o ganiza ional da ase s by in es iga ing hese chal-
lenges in he ield o inancial aud de ec ion. This was done
by de i ing DPs o an inno a i e syn he ic da a-sha ing
ecosys em ha allows inancial ins i u ions o exchange
inancial ansac ion da a while p o ec ing clien p i acy
and lea ning e ec i ely om his mul i-ins i u ional da a.
To c ea e his a i ac , we ollowed he p ocess o DSRM
(Pe e s e al., 2007), wi h his pape co e ing ou “design-
implemen -e alua e” cycles. S a ing wi h he p oblem iden-
i ica ion ou s udy con ibu es o desc ip i e knowledge
conce ning he p oblem space by iden i ying da a sca ci y
in combina ion wi h he inabili y o sha e da a due o p i-
acy p o ec ion as a majo hu dle o inancial ins i u ions,
alida ing he exis ing esea ch on c oss-o ganiza ional
aud de ec ion collabo a ion wi hin inancial se ices
(Abdul Salam e al., 2024; Kong e al., 2024). Du ing he
explo a ion o he solu ion, space syn he ic da a sha ing
Fig. 10 Pe o mance gain pe indi idual bank and pe o mance gain by ins i u ion size
Fig. 11 Pe o mance (a g pe bank) by pe cen age o pa icipa ing
ins i u ions

Elec onic Ma ke s (2025) 35:7 Page 21 o 28 7
was iden i ied as an unde explo ed solu ion o ackle da a
sca ci y in inancial aud de ec ion ex ending he li e a u e
on c oss-o ganiza ional collabo a ion in he ield (Cha e -
jee e al., 2024). Fu he mo e, he explo a ion o syn he ic
da a o allow p i acy-complian da a sha ing as well as ou
expe imen a ion on mul i-o ganiza ional syn he ic da a du -
ing mul iple “design-implemen -e alua e” cycles eaches
beyond inancial se ices and add esses signi ican chal-
lenges in he ealm o da a ecosys ems (B ée e al., 2024).
Mo eo e , ou esea ch ex ends beyond s udies ha simply
ou line he equi emen s o such a da a ecosys em (Immonen
e al., 2014). We alida e hese equi emen s and he de i ed
DPs h ough igo ous expe imen a ion on publicly a ailable
da ase s and h ough close collabo a ion wi h indus y pa -
ne s and expe s, ensu ing he p ac ical applicabili y and
obus ness o ou indings. Fu he mo e, by ex ending da a
ecosys em esea ch in o a less equen ly explo ed domain
(Cappiello e al., 2020), we a e able o alida e he applica-
bili y o exis ing knowledge and unco e new insigh s wi h
po en ial o gene aliza ion. We achie e his by de eloping
p esc ip i e knowledge and nascen heo y conce ning he
solu ion space, o e ing a se o DPs o designing a syn he ic
da a sha ing ecosys em and p o iding a i s ins an ia ion in
he o m o a pla o m a chi ec u e. To p o ide mo e de ailed
insigh s in o his solu ion space, addi ional key indings a e
encapsula ed in Table9, clus e ed by key a eas which we
deduc i ely de i ed a pos e io i om ou s udy.
As shown in Table9 unde he gene a ion dimension, we
con ibu e o he li e a u e on syn he ic da a gene a ion in
mul iple ways. Fi s , we iden i ied he necessi y o a s ic ly
sepa a ed local laye (whe e eal da a is ans o med) and
a global laye (whe e da a is sha ed). Second, we ans e
exis ing algo i hms o a new se up including c oss-o gan-
iza ional da a wi h a complex da a s uc u e and compa e
hei pe o mance on a p edic ion ask (Pa ha e e al., 2023)
iden i ying TVAE as he mos pe o man algo i hm o syn-
he ic inancial da a gene a ion while s ill showing su icien
p i acy. Thi d, we ex end he esea ch on he gene a ion
se up by consolida ing di e en aining schemes om mul-
iple sou ces (Eile sen e al., 2021; Fan e al., 2022; Ki an &
Kuma , 2024) and compa ing hem o each o he , iden i ying
aining on da a sub-clus e s as he mos bene icial se up.
Mo ing o wa d o aining models based on syn he ic
da a, as shown in Table9 unde he p edic ion dimension,
we ex end he li e a u e which o en looks a syn he ic da a
gene a ion pe o mance sepa a ely bu p o ides li le guid-
ance on how he gene a ed da a is bes used in a da a eco-
sys em (Danka e al., 2022). Ou esea ch u he shows
ha a mix u e o syn he ic and eal da a is mos use ul when
combined; howe e , he exac mix-in pe cen age is highly
o ganiza ion and con ex -speci ic. Mo eo e , we demon-
s a ed ha using pu ely syn he ic da a can s ill be bene icial
o playe s wi h no access o eal da a; howe e , adjus men s
need o be made o he composi ion o he da a by a i icially
ebalancing i .
As can be seen in Table9’s ecosys em dimension, ou
esea ch in es iga es he complexi ies o da a ecosys ems,
analyzing how he incen i es o pa icipa ion a ec pe -
o mance ou comes ac oss a ious sizes o ins i u ions.
This analysis also places ou indings in he con ex o he
esea ch by Gelhaa and O o (2020) abou he ini ial chal-
lenges encoun e ed wi hin da a ecosys ems. By implemen -
ing design in e en ions ha clea ly a icula e pe o mance
bene i s and acili a e he in eg a ion o ex e nal collabo a-
o s, ou esea ch subs an ia es he ecosys em’s capaci y o
o e come hese ea ly hu dles. Fu he , ou empi ical e i-
dence sugges s ha e en pa ial pa icipa ion in he da a eco-
sys em can lead o subs an ial pe o mance imp o emen s,
he eby a i ming he ecosys em’s ope a ional easibili y and
enhancing i s a ac i eness o po en ial pa icipan s.
In he las dimension in Table9, we demons a e he
gene alizabili y o ou de i ed design knowledge beyond
a single use case and applica ion a ea. This was done in
wo ways. Fi s , h ough alida ion wi h expe s om he
ield in academia and he p i a e sec o . Second, h ough
pe o mance e alua ion in wo inancial se ices domains
and h ee da ase s which equi ed da a sha ing wi h p i acy
es ic ions. While pe o mance gains migh seem insigni i-
can , small changes in aud de ec ion a e can ha e majo
implica ions on inancial ins i u ions (Le i, 1998). Thus, ou
esea ch no only con i ms he ele ance o ou DPs and sys-
em a chi ec u e bu also se s he s age o hei applica ion
beyond he immedia e con ex o inancial ansac ions, sug-
ges ing a bluep in o ex ending beyond inancial se ices
o o he domains whe e da a needs o be sha ed wi h p i acy
es ic ions (Susha e al., 2019).
Fo p ac i ione s, ou con ibu ion is wo- old: Fo man-
age s and decision-make s, we demons a e he alue o
syn he ic da a-sha ing ecosys ems ha allow bo h la ge
and small ins i u ions o secu ely collabo a e on da a while
ensu ing p i acy. This app oach is pa icula ly ele an in
indus ies wi h complex, highly sensi i e da a, such as inan-
cial se ices, whe e da a ecosys ems do no eme ge o gani-
cally and equi e ca e ul planning allowing o sha ed alue
p oposi ions and se ices (Adne , 2017; Immonen e al.,
2014). Fu he mo e, ou amewo k add esses egula o y
equi emen s on da a p i acy and ou esul s sugges a
obus ounda ion o scaling and sus aining p i acy- ocused
da a ecosys ems. Fo sys em a chi ec s, we ou line a se o
DPs ha guide p ac i ione s in s uc u ing he a chi ec u e
o hese ecosys ems. These p inciples assis in selec ing
sui able syn he ic da a gene a ion me hods, implemen ing
mechanisms o da a quali y assu ance, and in eg a ing da a
o enhance AI model pe o mance. By ocusing on hese
co e a eas, ou con ibu ion p o ides a chi ec s wi h ac ion-
able guidance owa d building secu e and esilien syn he ic
Elec onic Ma ke s (2025) 35:77 Page 22 o 28
Table 9 Summa y o insigh s gene a ed in he ou design-implemen -e alua e cycles
* Pe o mance e alua ion based on ROC AUC sco es
A ea o insigh Focus Ac i i ies Reasoning
Gene a ion This dimension iden i ies he mos sui able syn he ic
da a gene a ion algo i hm and he op imal se up o
gene a ing syn he ic da a o enhance pe o mance
- Pe o mance compa ison o popula syn he ic da a
gene a ion algo i hms on eal-wo ld inancial ansac-
ion da a (ou pe o mance o TVAE by 49.8%* com-
pa ed o nex bes model)
- E alua ion o he op imal aining se up when gene a -
ing syn he ic da a (ou pe o mance o p e- ained
class-sepa a ed models by 2.0%* compa ed o nex
bes aining scheme)
The analyses lead o he selec ion o a modula , algo-
i hm-agnos ic app oach ha clea ly sepa a es be ween
local and global laye s. Fu he , op imal da a in eg a ion
s a egies we e iden i ied b oadening he ecosys em’s
applicabili y and enhancing i s pe o mance po en ial
P edic ion This dimension ocuses on he speci ics o aining
se ups o syn he ic da a gene a ion, especially in
con ex s wi h imbalanced classes
- E alua ion o he op imal mix-in pe cen age be ween
eal and syn he ic da a on he ins i u ional le el
- Pe o mance e alua ion o pu ely syn he ic da a using
di e en deg ees o o e sampling (ou pe o mance o
da a wi h 10% o e sampling by 7.5%* compa ed o no
o e sampled baseline)
These in es iga ions acili a e he iden i ica ion o ailo ed
aining app oaches ha accommoda e he speci ic
needs o he ecosys em’s pa icipan s, ensu ing he
gene a ion o high-quali y, u ili y-maximizing p edic-
ion models
Ecosys em This dimension assesses he ecosys em’s balance in
e ms o pa icipa ion incen i es ac oss ins i u ions o
a ying sizes and he e ec o pa ial pa icipa ion
- Analysis o pe o mance gain pe inancial ins i u-
ion and he ela ionship be ween ins i u ion size and
pe o mance gain (nega i e co ela ion o − 0.09)
- E alua ion o he ecosys em’s pe o mance wi h a y-
ing numbe s o pa icipa ing ins i u ions (signi ican
pe o mance gain wi h all ecosys em sizes)
The indings indica e a p opo ional bene i ac oss he
ecosys em, highligh ing pa icula ad an ages o
smalle ins i u ions. Fu he mo e, e en pa ial eco-
sys em pa icipa ion yields subs an ial pe o mance
imp o emen s. Thus, ensu ing he sys em's iabili y and
a ac i eness
Gene alizabili y The dimension o gene alizabili y aims o alida e he
c ea ed design knowledge and sys em a chi ec u e
ac oss di e se con ex s and expe s’ pe spec i es
- Semi-s uc u ed in e iews wi h expe s om inancial
ins i u ions and academic expe s o alida e he c e-
a ed DPs and sys em a chi ec u e
- T ans e o he syn he ic da a sha ing se up o new
con ex s like aud de ec ion (imp o ed pe o mance
by 3.6%*) and money launde ing de ec ion (imp o ed
pe o mance by 6.6%*) using simula ion-based da a-
se s
The e alua ion con i med he ele ance and applicabil-
i y o he design knowledge ac oss mul iple inancial
se ices domains
Elec onic Ma ke s (2025) 35:7 Page 23 o 28 7
da a-sha ing ecosys ems. This amewo k, he e o e, se es
as a bluep in o u u e sys em designe s wo king wi hin
egula ed en i onmen s whe e da a p i acy and AI pe o -
mance a e essen ial.
Limi a ions and u u e esea ch oppo uni ies can be iden-
i ied ac oss ou ou key a eas o insigh . Rega ding da a
gene a ion, he cu en s udy was cons ained by he a ail-
able da a, which p e en ed he conside a ion o ad anced
g aph-based syn he ic da a gene a ion me hods such as
T ansGAN (X. Wang & Yang, 2024). Addi ionally, while
p i acy was es ed, i was no ully gua an eed by he mod-
els used, highligh ing he need o u u e esea ch on he
e ec i eness o di e en ially p i a e syn he ic da a gene a-
ion me hods such as PATEGAN (Jo don e al., 2018) in
a syn he ic da a ecosys em. F om a p edic ion s andpoin ,
u he in es iga ion is equi ed o de e mine how models
can be aligned when da a schemas—and hus he syn he ic
da a—di e be ween ins i u ions. Mo eo e , he design o
an e ec i e back- es ing mechanism o ensu e he ecosys-
em’s p edic i e pe o mance should be explo ed. On he
ecosys em le el, addi ional esea ch is necessa y o explo e
ecosys em usage incen i es, building on he wo k by (Gel-
haa e al., 2021), which was beyond he scope o his pape .
Finally, while his s udy was limi ed o inancial se ices due
o esou ce cons ain s, u u e esea ch should explo e he
applicabili y o he de ined DPs beyond his domain, es ing
hei gene al applicabili y.
Conclusion
Based on he need o inc eased da a a ailabili y o os e
economic g ow h, his pape p o ides he design and e alu-
a ion o a syn he ic da a-sha ing ecosys em o inancial
ins i u ions unde p i acy cons ain s. The main con ibu ion
lies in p o iding guidance on how o ain models based on
sha ed da a. By o mula ing a se o DPs, p ac ical insigh s,
and p o o ype es ing, i e a i e design cycles we e used o
p o ide a obus amewo k o cons uc ing a da a ecosys-
em ha le e ages syn he ic da a. Each DP, om ensu ing
da a quali y and enhancing adap abili y h ough ans o ma-
ion and esampling o os e ing us among ecosys em pa -
icipan s and acili a ing egula o y access o syn he ic da a,
ex ends exis ing esea ch on syn he ic da a sha ing and gen-
e a ion, pa icula ly in he con ex o inancial ansac ion
da a. Fo p ac ice, ou example ins an ia ion and codebase
can be used as a e e ence a chi ec u e o u u e ins an ia-
ions. We no only add ess he iden i ied need o an e icien ,
p i acy-p ese ing inancial da a ecosys em bu also se a
ounda ion o u u e explo a ion in b oade domains whe e
da a sha ing unde p i acy es ic ions is pa amoun . Thus,
his con ibu ion o e s guidance o o e coming echnical,
us - ela ed, and egula o y challenges in da a ecosys ems,
unlocking he po en ial o da a-d i en inno a ion and u u e
economic de elopmen .
Acknowledgemen s The au ho s exp ess hei g a i ude o he Union-
Bank o he Philippines o hei aluable collabo a ion and he p o i-
sion o key insigh s ha con ibu ed o his esea ch.This esea ch p o-
jec was unded by he S . Gallen Symposium and he Ge man Fede al
Minis y o Educa ion and Resea ch (BMBF) wi hin he “Inno a ions
o Tomo ow’sP oduc ion, Se ices, and Wo k” P og am ( unding
numbe 02K23A001) which is managed by he P ojec Managemen
Agency Ka ls uhe (PTKA). The au ho s a e esponsible o hecon en
o his publica ion.
Funding Open access unding p o ided by Uni e si y o S . Gallen.
Da a A ailabili y The da ase s used du ing he cu en s udy a e a ail-
able in he Kaggle eposi o ies: h ps:// www. kaggle. com/c/ ieee- aud-
de ec ion,h ps:// www. kaggle. com/ da as e s/ eal m an2019/ ibm- ans
ac io ns- o - an i- money- laund e ing- aml, h ps:// www. kaggle. com/ da as
e s/ eal m an2019/ c edi - ca d- ans ac io ns.
Decla a ions
Compe ing In e es s The au ho s decla e ha hey ha e no con lic
o in e es .
Open Access This a icle is licensed unde a C ea i e Commons A i-
bu ion 4.0 In e na ional License, which pe mi s use, sha ing, adap a-
ion, dis ibu ion and ep oduc ion in any medium o o ma , as long
as you gi e app op ia e c edi o he o iginal au ho (s) and he sou ce,
p o ide a link o he C ea i e Commons licence, and indica e i changes
we e made. The images o o he hi d pa y ma e ial in his a icle a e
included in he a icle's C ea i e Commons licence, unless indica ed
o he wise in a c edi line o he ma e ial. I ma e ial is no included in
he a icle's C ea i e Commons licence and you in ended use is no
pe mi ed by s a u o y egula ion o exceeds he pe mi ed use, you will
need o ob ain pe mission di ec ly om he copy igh holde . To iew a
copy o his licence, isi h p:// c ea i eco mmons. o g/ licen ses/ by/4. 0/.
Re e ences
Abbas, A. E., Agaha i, W., an de Ven, M., Zuide wijk, A., & de
Reu e , M. (2021). Business da a sha ing h ough da a ma ke -
places: A sys ema ic li e a u e e iew. Jou nal o Theo e ical and
Applied Elec onic Comme ce Resea ch, 16(7), 7. h ps:// doi. o g/
10. 3390/ j ae 16070 180
Abdul Salam, M., Fouad, K. M., Elbably, D. L., & Elsayed, S. M.
(2024). Fede a ed lea ning model o c edi ca d aud de ec-
ion wi h da a balancing echniques. Neu al Compu ing and
Applica ions, 36(11), 6231–6256. h ps:// doi. o g/ 10. 1007/
s00521- 023- 09410-2
Adne , R. (2017). Ecosys em as s uc u e: An ac ionable cons uc o
s a egy. Jou nal o Managemen , 43(1), 39–58. h ps:// doi. o g/
10. 1177/ 01492 06316 678451
Al-Hashedi, K. G., & Magalingam, P. (2021). Financial aud de ec ion
applying da a mining echniques: A comp ehensi e e iew om
2009 o 2019. Compu e Science Re iew, 40, 100402. h ps:// doi.
o g/ 10. 1016/j. cos e . 2021. 100402
Al man, E., Blanuša, J., on Niede häuse n, L., Eg essy, B., Anghel,
A., & A asu, K. (2024). Realis ic syn he ic inancial ansac-
ions o an i-money launde ing models (No. a Xi : 2306. 16424).
a Xi . h ps:// doi. o g/ 10. 48550/ a Xi . 2306. 16424
Elec onic Ma ke s (2025) 35:77 Page 24 o 28
Al man, E. R. (2019). Syn hesizing c edi ca d ansac ions (No.
a Xi : 1910. 03033). a Xi . h ps:// doi. o g/ 10. 48550/ a Xi .
1910. 03033
As ow, K. (2021). The ole o indi iduals in he da a ecosys em: Cu -
en deba es and conside a ions o indi idual da a p o ec ion
and da a igh s in he U.S. Fede al Rese e Bank o San F an-
cisco. h ps:// p i a cysec u i y acade my. com/ wp- con e n / uploa ds/
2021/ 05/ The- Role- o - Indi iduals- in- he- Da a- Ecosy s em. pd .
Accessed 9 Ma 2023.
Asse a, S. (2020). Gene a ing syn he ic da a in inance: Oppo uni ies,
challenges and pi alls. SSRN Elec onic Jou nal. h ps:// doi. o g/
10. 2139/ ss n. 36342 35
Au na, N. F., Hossain, M. D., Taenaka, Y., & Kadobayashi, Y. (2023).
Fede a ed lea ning-based c edi ca d aud de ec ion: Pe o -
mance analysis wi h sampling me hods and deep lea ning algo-
i hms. 2023 IEEE In e na ional Con e ence on Cybe Secu i y
and Resilience (CSR), 2023, 180–186. h ps:// doi. o g/ 10. 1109/
CSR57 506. 2023. 10224 978
Au io, E., spsampsps Thomas, L. D. W. (2014). Inno a ion ecosys ems:
Implica ions o inno a ion managemen ? In M. Dodgson, D. M.
Gann, spsampsps N. Phillips (Eds.), The Ox o d Handbook o
Inno a ion Managemen (p. 0). Ox o d Uni e si y P ess. h ps://
doi. o g/ 10. 1093/ ox o dhb/ 97801 99694 945. 013. 012
Awosika, T., Shukla, R. M., & P anggono, B. (2024). T anspa ency
and p i acy: The ole o explainable AI and ede a ed lea ning in
inancial aud de ec ion. IEEE Access, 12, 64551–64560. h ps://
doi. o g/ 10. 1109/ ACCESS. 2024. 33945 28. IEEE Access.
Baabdullah, T., Alzah ani, A., Rawa , D. B., & Liu, C. (2024). E i-
ciency o ede a ed lea ning and blockchain in p ese ing p i acy
and enhancing he pe o mance o c edi ca d aud de ec ion
(CCFD) sys ems. Fu u e In e ne , 16(6), 6. h ps:// doi. o g/ 10.
3390/ i160 60196
Bagad, P., Mi a, S., Dhamnani, S., Sinha, A. R., Gau am, R., &
Khanna, H. (2021). Da a-sha ing economy: Value-addi ion om
da a mee s p i acy. P oceedings o he 14 h ACM In e na ional
Con e ence on Web Sea ch and Da a Mining, 1105–1108. h ps://
doi. o g/ 10. 1145/ 34379 63. 34417 12
Baske ille, R., Baiye e, A., G ego , S., He ne , A., & Rossi, M.
(2018). Design science esea ch con ibu ions: Finding a bal-
ance be ween a i ac and heo y. Jou nal o he Associa ion o
In o ma ion Sys ems, 19(5). h ps:// aisel. aisne . o g/ jais/ ol19/
iss5/3. Accessed 13 June 2023.
Baue , A., T app, S., S enge , M., Leppich, R., Koune , S., Leznik,
M., Cha d, K., & Fos e , I. (2024). Comp ehensi e explo a-
ion o syn he ic da a gene a ion: Asu ey (No. a Xi : 2401.
02524). a Xi . h ps:// a xi . o g/ abs/ 2401. 02524. Accessed 15
Aug 2024.
Benchaji, I., Douzi, S., & Ouahidi, B. E. (2021). C edi ca d aud
de ec ion model based on LSTM ecu en neu al ne wo ks.
Jou nal o Ad ances in In o ma ion Technology, 12(2), 113–118.
h ps:// doi. o g/ 10. 12720/ jai . 12.2. 113- 118
Bian, K., & Zheng, H. (2023). FedA g-DWA: A no el algo i hm o
enhanced aud de ec ion in ede a ed lea ning en i onmen .
2023 4 h In e na ional Con e ence on Big Da a, A i icial In el-
ligence and In e ne o Things Enginee ing (ICBAIE), 13–17.
h ps:// doi. o g/ 10. 1109/ ICBAI E59714. 2023. 10281 317
Blake, M., McWa e s, J., & Galaski, R. (2019). The nex gene a ion o
da a-sha ing in inancial se ices (p. 33) [Whi e Pape ]. Wo ld
Economic Fo um. h ps:// www2. deloi e. com/ con e n / dam/ Deloi
e/ lu/ Docum en s/ inan cial- se i ces/ lu- nex - gene a ion- da a-
sha i nging- inan cial- se i ces. pd . Accessed 29 Jan 2023.
Bonabeau, E. (2002). Agen -based modeling: Me hods and echniques
o simula ing human sys ems. P oceedings o he Na ional
Academy o Sciences, 99(suppl_3), 7280–7287. h ps:// doi. o g/
10. 1073/ pnas. 08208 0899
B ée, T., Ka ge , E., & Ahlemann, F. (2024). Shaping he u u e o da a
ecosys em esea ch—Wha is s ill missing? IEEE Access, 12,
103162–103175. IEEE Access. h ps:// doi. o g/ 10. 1109/ ACCESS.
2024. 34329 69
B odsky, L., & Oakes, L. (2017). Da a sha ing and open banking.
McKinsey. h ps:// www. mckin sey. com/ ~/ media/ McKin sey/ Indus
ies/ Finan cial% 20Se ices/ Ou % 20Ins igh s/ Da a% 20sha ing%
20and% 20open% 20ban king/ Da a- sha i ng- and- open- banki ng. pd .
Accessed 26 Feb 2024.
Bun, M., Des on aines, D., Dwo k, C., Nao , M., Nissim, K., Ro h, A.,
Smi h, A., S einke, T., Ullman, J., & Vadhan, S. (2021). S a is-
ical in e ence is no a p i acy iola ion. h ps:// di e en i alp i
acy. o g/ in e ence- is- no -a- p i a cy- iola ion/. Accessed 14 Aug
2024.
Cappiello, C., Gal, A., Ja ke, M., & Reho , J. (2020). Da a ecosys ems:
So e eign da a exchange among o ganiza ions (Dags uhl Semi-
na 19391). DROPS-IDN/ 2/Documen /h ps:// doi. o g/ 10. 4230/
DagRep. 9.9. 66. h ps:// doi. o g/ 10. 4230/ DagRep. 9.9. 66
Chak a a y, S., Demi han, H., & Base , F. (2020). Fuzzy eg es-
sion unc ions wi h a noise clus e and he impac o ou lie s on
mains eam machine lea ning me hods in he eg ession se ing.
Applied So Compu ing, 96, 106535. h ps:// doi. o g/ 10. 1016/j.
asoc. 2020. 106535
Chand a, L., Seidel, S., & G ego , S. (2015). P esc ip i e knowledge in
IS esea ch: Concep ualizing design p inciples in e ms o ma e-
iali y, ac ion, and bounda y condi ions. 2015 48 h Hawaii In e -
na ional Con e ence on Sys em Sciences, 4039–4048. h ps:// doi.
o g/ 10. 1109/ HICSS. 2015. 485
Cha i ou, C., D agice ic, S., & Ga cez, A. d’A ila. (2021). Syn he ic
da a gene a ion o aud de ec ion using GANs (No. a Xi : 2109.
12546). a Xi . h p:// a xi . o g/ abs/ 2109. 12546. Accessed 11 Ma
2024.
Cha e jee, P., Das, D., & Rawa , D. B. (2024). Digi al win o c edi
ca d aud de ec ion: Oppo uni ies, challenges, and aud de ec-
ion ad ancemen s. Fu u e Gene a ion Compu e Sys ems, 158,
410–426. h ps:// doi. o g/ 10. 1016/j. u u e. 2024. 04. 057
Chen, Z., Van Khoa, L. D., Teoh, E. N., Nazi , A., Ka uppiah, E. K., &
Lam, K. S. (2018). Machine lea ning echniques o an i-money
launde ing (AML) solu ions in suspicious ansac ion de ec ion:
A e iew. Knowledge and In o ma ion Sys ems, 57(2), 245–285.
h ps:// doi. o g/ 10. 1007/ s10115- 017- 1144-z
Chen, H., G ossman, M., Sen, A., & Tsao, S.-F. (2023). Es ablish-
ing a FAIR, CARE, and e icien syn he ic heal h da a sha -
ing ecosys em o canada es ablishing a FAIR, CARE, and
e icien syn he ic heal h da a sha ing ecosys em o Canada.
IARIW-CIGI Con e ence on he Valua ion o Da a. h ps://
www. esea chga e. ne / publi ca ion/ 37544 6378_ Es ab lishi
ng_a_ FAIR_ CARE_ and_ E ic ien _ Syn h e ic_ Heal h_ Da a_
Sha i ng_ Ecosy s em_ o _ Canada_ Es ab lishi ng_a_ FAIR_
CARE_ and_ E ic ien _ Syn h e ic_ Heal h_ Da a_ Sha i ng_ Ecosy
s em_ o _ Canada.Accessed 17 Dec 2024
Choi, E., Biswal, S., Malin, B., Duke, J., S ewa , W. F., & Sun, J.
(2018). Gene a ing mul i-label disc e e pa ien eco ds using
gene a i e ad e sa ial ne wo ks (No. a Xi : 1703. 06490). a Xi .
h ps:// doi. o g/ 10. 48550/ a Xi . 1703. 06490
Cosma, S., Cosma, S., & Penne a, D. (2023). The ise o inancial se -
ices ecosys ems: Towa ds open banking pla o ms. In T. Walke ,
E. Nikbakh , & M. Kooli (Eds.), The Fin ech Dis up ion: How
Financial Inno a ion Is T ans o ming he Banking Indus y (pp.
191–213). Sp inge In e na ional Publishing. h ps:// doi. o g/ 10.
1007/ 978-3- 031- 23069-1_8
Dahmen, J., & Cook, D. (2019). SynSys: A syn he ic da a gene a ion
sys em o heal hca e applica ions. Senso s, 19(5), 5. h ps:// doi.
o g/ 10. 3390/ s1905 1181
Danka , F. K., Ib ahim, M. K., & Ismail, L. (2022). A mul i-dimen-
sional e alua ion o syn he ic da a gene a o s. IEEE Access, 10,