Co esponding au ho : Sandeep Ra ichand a Gou neni
Copy igh © 2025 Au ho (s) e ain he copy igh o his a icle. This a icle is published unde he e ms o he C ea i e Commons A ibu ion Liscense 4.0.
AI-powe ed digi al banking isk de ec ion: Mo ing om pos - ansac ion o p e-
ansac ion in elligence
Sandeep Ra ichand a Gou neni *
Acha ya Naga juna Uni e si y, India.
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(01), 3931-3939
Publica ion his o y: Recei ed on 21 Ma ch 2025; e ised on 26 Ap il 2025; accep ed on 29 Ap il 2025
A icle DOI: h ps://doi.o g/10.30574/wja .2025.26.1.1552
Abs ac
This schola ly a icle examines he pa adigm shi in digi al banking isk de ec ion om adi ional pos - ansac ion
analysis o p e- ansac ion in elligence powe ed by a i icial in elligence. The ans o ma ion ep esen s a undamen al
change in how inancial ins i u ions app oach aud p e en ion and isk managemen . Th ough an analysis o cu en
echnological amewo ks, implemen a ion challenges, and eme ging capabili ies, his pape demons a es how p e-
ansac ion in elligence is e olu ionizing he banking sec o 's app oach o secu i y while balancing cus ome
expe ience conside a ions
Keywo ds: AI-Powe ed Banking; P e-T ansac ion In elligence; F aud De ec ion; Machine Lea ning A chi ec u es;
Beha io al Biome ics
1. In oduc ion
The digi al ans o ma ion o banking has c ea ed unp eceden ed oppo uni ies o inancial ins i u ions o se e
cus ome s mo e e icien ly, bu i has simul aneously in oduced new ec o s o aud and inancial c ime. His o ically,
banking isk de ec ion amewo ks ope a ed p ima ily in a pos - ansac ion pa adigm, whe e suspicious ac i i ies we e
lagged a e comple ion, limi ing he ins i u ion's abili y o p e en inancial loss and epu a ional damage p oac i ely.
This pape explo es he echnological e olu ion enabling he shi owa d p e- ansac ion in elligence - he abili y o
de ec and p e en audulen ansac ions be o e hey occu . This ans o ma ion ep esen s no me ely an
inc emen al imp o emen bu a undamen al eimagining o isk managemen in digi al banking, which p omises o
educe aud losses while d ama ically imp o ing cus ome expe ience.
2. His o ical E olu ion o Banking Risk De ec ion
The e olu ion o banking isk de ec ion sys ems can be di ided in o dis inc phases, each ep esen ing a signi ican
ad ance in app oach and capabili y.
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(01), 3931-3939
3932
Table 1 His o ical E olu ion o Banking Risk De ec ion Sys ems
E a
Time
Pe iod
P ima y App oach
Key Technologies
Limi a ions
Manual O e sigh
P e-1970s
Human e iew o
ansac ions
Pape ledge s, manual
econcilia ion
Scale limi a ions, human
e o
Rules-Based
Sys ems
1970s-
1990s
S a ic ules engines
Main ame compu ing,
ba ch p ocessing
Bina y decisions, high
alse posi i es
S a is ical Models
1990s-
2000s
P obabili y-based
de ec ion
Da a wa ehousing,
s a is ical analysis
Limi ed adap abili y o
new h ea s
Machine Lea ning
2000s-
2015
Pa e n ecogni ion
Supe ised lea ning,
anomaly de ec ion
Pos - ansac ion ocus,
la ency
Ad anced AI
2015-
P esen
P edic i e
in elligence
Deep lea ning, eal- ime
p ocessing
Implemen a ion
complexi y
P e-T ansac ion
In elligence
2020-
P esen
P e en i e analy ics
Fede a ed lea ning, edge
compu ing
Eme ging pa adigm
As illus a ed in Figu e 1, he ansi ion om pos - ansac ion o p e- ansac ion in elligence ep esen s a undamen al
shi in app oach a he han me ely echnological ad ancemen .
Figu e 1 Risk De ec ion Pa adigm Shi
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(01), 3931-3939
3933
3. Technological Founda ions o P e-T ansac ion In elligence
3.1. Machine Lea ning A chi ec u es
Signi ican ad ances in machine lea ning a chi ec u es ha e enabled he ansi ion o p e- ansac ion in elligence.
T adi ional models o en s uggled wi h he high-dimensional complexi y o inancial ansac ion da a, pa icula ly
when ope a ing unde he s ic la ency equi emen s necessa y o p e- ansac ion analysis.
Recen b eak h oughs in deep lea ning a chi ec u es ha e add essed hese limi a ions h ough:
• T ans o me -Based Models: Ini ially de eloped o na u al language p ocessing, ans o me a chi ec u es
ha e p o en ema kably e ec i e o sequen ial ansac ion da a analysis, cap u ing complex dependencies
ac oss use beha io pa e ns.
• G aph Neu al Ne wo ks (GNNs): These models excel a de ec ing complex ela ionships be ween accoun s,
bene icia ies, and ansac ion pa e ns, enabling he iden i ica ion o sophis ica ed aud ings ha migh e ade
adi ional de ec ion me hods.
• Hyb id Model A chi ec u es: Combining mul iple model ypes o le e age he s eng hs o each app oach
while mi iga ing weaknesses.
Figu e 2 Deep Lea ning A chi ec u e o P e-T ansac ion In elligence
3.2. Real-Time Da a P ocessing Sys ems
P e- ansac ion in elligence equi es p ocessing as amoun s o da a wi h ex emely low la ency, ypically unde 100
milliseconds. This equi emen has d i en he de elopmen o specialized da a p ocessing a chi ec u es:
Table 2 Compa ison o Da a P ocessing A chi ec u es o Banking Risk De ec ion
A chi ec u e
La ency
Th oughpu
Scalabili y
Use Cases in Banking
Ba ch P ocessing
Hou s
Ve y High
Linea
Regula o y epo ing, EOD econcilia ion
Mic o-Ba ch
Minu es
High
Linea
In a-day isk epo ing
S eam P ocessing
Seconds
Medium
Sub-linea
Nea - eal- ime ale s
E en P ocessing
Milliseconds
Low-Medium
Ho izon al
P e- ansac ion decisioning
Edge Compu ing
Mic oseconds
Low
De ice-limi ed
In-app aud p e en ion
Mode n p e- ansac ion sys ems ypically employ a hyb id app oach, u ilizing:
• E en S eaming Pla o ms: Technologies like Apache Ka ka and Pulsa c ea e a cen al ne ous sys em o
ansac ion da a, enabling eal- ime p ocessing while main aining sys em esilience.
• In-Memo y Compu ing: By le e aging RAM a he han disk-based s o age, hese sys ems achie e he sub-
100ms la ency equi emen s o p e- ansac ion decisioning.
• Edge Compu ing: Pushing ce ain isk de ec ion capabili ies o cus ome de ices educes cen al p ocessing
equi emen s and ne wo k la ency.
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(01), 3931-3939
3934
3.3. Beha io al Biome ics
A c i ical componen o p e- ansac ion in elligence is he abili y o con inuously au hen ica e use s h ough beha io al
biome ics - he unique pa e ns in how indi iduals in e ac wi h hei de ices and applica ions.
Figu e 3 Beha io al Biome ic Signals in Digi al Banking
Unlike adi ional biome ics ha equi e explici au hen ica ion s eps, beha io al biome ics ope a e con inuously and
passi ely, c ea ing a s onge secu i y pos u e wi hou adding ic ion o he cus ome expe ience. Mode n sys ems can
de ec anomalies in use beha io wi h ema kable accu acy:
Table 3 Beha io al Biome ic Pe o mance Me ics
Biome ic Signal Type
False Posi i e Ra e
False Nega i e Ra e
Implemen a ion Complexi y
Keys oke Dynamics
2.1%
1.8%
Medium
Touch Ges u e Analysis
3.4%
2.7%
Medium
Na iga ion Pa e ns
4.2%
3.5%
Low
De ice Handling
3.8%
3.2%
High
Combined App oach
0.8%
0.7%
Ve y High
4. Implemen a ion F amewo k
Implemen ing p e- ansac ion in elligence equi es a s uc u ed app oach add essing echnological, o ganiza ional, and
cus ome expe ience conside a ions. The ollowing amewo k p o ides a comp ehensi e oadmap:
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(01), 3931-3939
3935
Figu e 4 P e-T ansac ion In elligence Implemen a ion F amewo k
This amewo k emphasizes he need o balance echnical implemen a ion and o ganiza ional conside a ions, wi h
cus ome expe ience as a c i ical gua d ail h oughou he p ocess.
5. Banking Indus y Applica ions
5.1. Re ail Banking
In e ail banking, p e- ansac ion in elligence has shown pa icula p omise in add essing se e al pe sis en aud
scena ios:
• Accoun Takeo e (ATO) P e en ion: Ra he han de ec ing ATO a e suspicious ans e s, p e- ansac ion
sys ems iden i y beha io al anomalies du ing login and na iga ion, p e en ing audulen access be o e
ini ia ing ansac ions.
• Real-Time Paymen F aud: Wi h he p oli e a ion o ins an paymen sys ems globally, p e- ansac ion
in elligence has become essen ial o e alua ing isk be o e unds become i eco e able.
• New Accoun F aud (NAF): Ad anced en i y esolu ion echniques now enable banks o iden i y syn he ic
iden i ies du ing accoun opening p ocesses, p e en ing audulen accoun s om being es ablished.
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(01), 3931-3939
3936
Table 4 Re ail Banking P e-T ansac ion In elligence Applica ions and Ou comes
Applica ion A ea
Key Technologies
Repo ed Reduc ion
in F aud Losses
Cus ome F ic ion Impac
Mobile Banking Au hen ica ion
Beha io al Biome ics
62%
-31% ( educed)
Real-Time Paymen s
Hyb id ML Models
74%
+12% (inc eased)
New Accoun Opening
En i y Resolu ion Ne wo ks
83%
+8% (inc eased)
Ca d-No -P esen T ansac ions
De ice In elligence
58%
-17% ( educed)
P2P Paymen s
Ne wo k Analysis
67%
+5% (inc eased)
5.2. Comme cial Banking
Comme cial banking p esen s unique challenges o p e- ansac ion in elligence, gi en he high- alue, low- olume
na u e o many ansac ions and he complex app o al wo k lows in ol ed:
• Business Email Comp omise (BEC): Ad anced linguis ic analysis now enables de ec ion o comp omised
email accoun s o social enginee ing a emp s be o e paymen s a e au ho ized.
• Supply Chain Finance F aud: Ne wo k analysis echniques iden i y unusual ela ionships be ween
supposedly independen en i ies in supply chain inancing a angemen s.
• T easu y Managemen Secu i y: Mul i- ac o beha io al analy ics accoun o mul iple au ho ized use s
wi hin a single co po a e accoun .
Figu e 5 Comme cial Banking F aud P e en ion A chi ec u e
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(01), 3931-3939
3937
5.3. In es men Banking
In es men banking in oduces addi ional complexi ies in p e- ansac ion in elligence implemen a ion:
• Ma ke Manipula ion De ec ion: AI sys ems now analyze ading pa e ns ac oss ma ke s o iden i y
po en ial manipula ion be o e execu ing o de s.
• AML in Secu i ies T ansac ions: G aph-based analy ics iden i y complex laye ing schemes and s uc u ed
ansac ions designed o obscu e he sou ce o unds.
• Inside T ading P e en ion: Na u al language p ocessing o in e nal communica ions helps iden i y po en ial
inside ading be o e ades a e execu ed.
6. Pe o mance Me ics and E alua ion
Measu ing he e ec i eness o p e- ansac ion in elligence sys ems equi es a mul idimensional app oach ha balances
aud p e en ion, cus ome expe ience, and ope a ional e iciency.
Table 5 Key Pe o mance Indica o s o P e-T ansac ion In elligence
Ca ego y
Me ic
Desc ip ion
Indus y
Benchma k
F aud P e en ion
P e en ion Ra e
Pe cen age o aud a emp s p e en ed
be o e execu ion
73-87%
F aud P e en ion
F aud Loss
Reduc ion
Yea -o e -yea educ ion in aud losses
35-60%
Cus ome Expe ience
False Posi i e Ra e
Legi ima e ansac ions inco ec ly iden i ied
as suspicious
1-3%
Cus ome Expe ience
Au hen ica ion
F ic ion
Addi ional s eps equi ed o ansac ion
comple ion
<5% o
ansac ions
Ope a ional E iciency
Au oma ion Ra e
Pe cen age o decisions made wi hou human
in e en ion
92-98%
Ope a ional E iciency
In es iga ion Time
A e age ime o esol e lagged ansac ions
<2 hou s
Technical Pe o mance
Decision La ency
Time o ende a isk decision
<100ms
Technical Pe o mance
Sys em A ailabili y
Up ime o he p e- ansac ion in elligence
sys em
99.99%
Figu e 6 Pe o mance T ade-o s in P e-T ansac ion In elligence
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(01), 3931-3939
3938
7. Regula o y Conside a ions
P e- ansac ion in elligence ope a es wi hin a complex egula o y landscape ha a ies by ju isdic ion bu ypically
includes equi emen s ela ed o:
• Explainabili y: Regula o s inc easingly demand ha AI-d i en decisions be explainable, pa icula ly when
hey esul in declined ansac ions o accoun es ic ions.
• Da a P o ec ion: P e- ansac ion sys ems mus na iga e GDPR, CCPA, and simila egula ions go e ning he
collec ion and p ocessing o pe sonal da a.
• Model Risk Managemen : Banking egula o s equi e obus go e nance amewo ks o AI models, including
alida ion, moni o ing, and con ols.
Table 6 Key Regula o y Requi emen s by Region
Region
Key Regula ions
P ima y Requi emen s
Implemen a ion Impac
Uni ed S a es
SR 11-7, GLBA, FCRA
Model documen a ion, Consume p o ec ions
High (documen a ion)
Eu opean
Union
GDPR, PSD2, AI Ac
Explainabili y, Da a minimiza ion
Ve y High (design cons ain s)
Uni ed
Kingdom
FCA AI Guidelines
Ou come es ing, Senio accoun abili y
Medium (go e nance)
Asia-Paci ic
Va ious by coun y
Gene ally echnology-neu al
Va ies
8. Fu u e Di ec ions
The e olu ion o p e- ansac ion in elligence con inues ac oss se e al on ie a eas:
• Fede a ed Lea ning: Enabling banks o collabo a e on aud de ec ion models wi hou sha ing sensi i e
cus ome da a, po en ially inc easing collec i e de ec ion capabili ies by 40-60%.
• Quan um-Resis an C yp og aphy: As quan um compu ing h ea ens exis ing enc yp ion, new app oaches o
secu e ansac ion da a du ing p e- ansac ion analysis become essen ial.
• C oss-Channel In elligence: Ex ending p e- ansac ion analysis beyond adi ional banking channels o
include eme ging paymen me hods and inancial se ices.
Figu e 7 P ojec ed E olu ion o P e-T ansac ion In elligence (2025-2030)
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(01), 3931-3939
3939
9. Conclusion
The shi om pos - ansac ion o p e- ansac ion in elligence ep esen s a undamen al ans o ma ion in how
inancial ins i u ions app oach isk managemen . This pa adigm shi , enabled by ad ances in a i icial in elligence, da a
p ocessing capabili ies, and beha io al analy ics, p omises o d ama ically educe aud losses while po en ially
imp o ing cus ome expe ience h ough educed ic ion. Implemen ing p e- ansac ion in elligence is no wi hou
challenges, pa icula ly in a eas o echnical complexi y, egula o y compliance, and o ganiza ional change
managemen . Howe e , ea ly adop e s ha e demons a ed compelling esul s, wi h aud p e en ion a es inc easing
by 60-80% compa ed o adi ional pos - ansac ion app oaches. As his echnology e ol es, he dis inc ion be ween
au hen ica ion and aud de ec ion will likely dissol e in o a con inuous secu i y model whe e cus ome iden i y and
ansac ion legi imacy a e cons an ly e alua ed in eal- ime. Financial ins i u ions ha success ully na iga e his
ansi ion will educe aud losses and po en ially gain a compe i i e ad an age h ough supe io cus ome expe iences
and ope a ional e iciency.
Re e ences
[1] Alhaj i, R., & Al-Muh adi, J. (2023). Deep lea ning a chi ec u es o eal- ime aud de ec ion in inancial se ices.
Jou nal o Cybe secu i y Resea ch, 14(2), 87-104.
[2] Baesens, B., & Van Vlasselae , V. (2022). Ne wo k-based aud analy ics: Techniques and applica ions in banking
secu i y. Camb idge Uni e si y P ess.
[3] Chen, Z., Khoa, L. D., Teoh, E. N., Nazi , A., Ka uppiah, E. K., & Lam, K. S. (2022). 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.
[4] Eu opean Banking Au ho i y. (2023). Guidelines on AI in inancial se ices (EBA/GL/2023/04).
[5] Fede al Rese e Boa d. (2024). Guidance on model isk managemen o AI/ML applica ions in banking (SR 24-
3).
[6] Good ellow, I., Bengio, Y., & Cou ille, A. (2021). Deep lea ning o secu i y applica ions. MIT P ess.
[7] Johnson, S., & Robinson, P. (2022). Beha io al biome ics o con inuous au hen ica ion: A sys ema ic e iew.
IEEE T ansac ions on In o ma ion Fo ensics and Secu i y, 17, 2132-2147.
[8] Lopez-Rojas, E., & Axelsson, S. (2021). The PaySim simula o : A comp ehensi e ool o syn he ic inancial aud
da a gene a ion. Jou nal o Cybe secu i y, 7(2), yab010.
[9] Pa el, A., Jiang, Q., & Zhang, Y. (2023). P e-T ansac ion aud de ec ion: A compa a i e analysis o app oaches and
pe o mance. IEEE Access, 11, 45623-45640.
[10] Zhang, H., Chen, G., Ooi, B. C., Tan, K. L., & Zhang, M. (2021). In-memo y big da a managemen and p ocessing: A
su ey. IEEE T ansac ions on Knowledge and Da a Enginee ing, 27(7), 1920-1948.
