Co esponding au ho : Ra i eja Gun upalli
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 License 4.0.
In elligen cloud ne wo king: Applying ai and ein o cemen lea ning o dynamic
a ic enginee ing, QoS op imiza ion and h ea de ec ion in so wa e-de ined cloud
a chi ec u es
Ra i eja Gun upalli *
Manage , Cloud Enginee ing, AnnA bo , Michigan, USA.
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 868-873
Publica ion his o y: Recei ed on 18 Ma ch 2025; e ised on 03 May 2025; accep ed on 06 May 2025
A icle DOI: h ps://doi.o g/10.30574/wja .2025.26.2.1520
Abs ac
Cloud ne wo ks o m he ounda ion o applica ions ha need dis ibu ed sys ems and equi e low la ency and op
pe o mance. The ising implemen a ion o SDN alongside mul i-cloud ne wo ks and edge sys ems has c ea ed
signi ican hu dles in managing ins an aneous a ic low pa e ns and secu i y h ea s oge he wi h ne wo k
conges ion. Con en ional ne wo k managemen using ules is unable o p ope ly con ol he la ge, di e se secu i y
h ea s p esen in cu en cloud en i onmen s. The in es iga ion demons a es how A i icial In elligence pu sues
op imiza ion o cloud ne wo k ope a ions by u ilizing ein o cemen lea ning (RL) and deep lea ning alongside g aph-
based models. The pape examines AI deploymen wi hin h ee undamen al ields - dynamic a ic enginee ing, Quali y
o Se ice op imiza ion, and secu i y-based anomaly de ec ion. The in eg a ion o ein o cemen lea ning agen s
demons a es hei abili y o pe o m adap i e eal- ime ne wo k a ic ou ing in combina ion wi h supe ised and
unsupe ised lea ning models, which p oduce conges ion p edic ions o QoS policy en o cemen . Ne wo k in usion
de ec ion has been success ully enhanced h ough he in eg a ion o AI sys ems in SDN-enabled cloud en i onmen s.
The applica ion o in elligen ne wo king o cloud se ice p o ide s is demons a ed h ough de ailed esea ch
in ol ing Mic oso Azu e and Google Cloud. The pape examines a ious p oduc ion challenges ega ding AI
deploymen in ne wo ks ha in ol e s abili y issues and explainabili y demands and equi e obus ness o ad e sa ial
inpu s and c oss-laye o ches a ion. Digi al se ice secu i y, high pe o mance, and adap abili y will ely on in elligen
ne wo king in as uc u e as cloud sys ems e ol e.
Keywo ds: Cloud ne wo king; AI-d i en a ic enginee ing; So wa e-de ined ne wo king (SDN); Rein o cemen
lea ning; QoS op imiza ion; DDoS de ec ion; Ne wo k anomaly de ec ion; In elligen ou ing; Conges ion con ol;
Au onomous ne wo ks
1. In oduc ion
The ne wo k ab ic ha suppo s cloud-na i e applica ions needs o p og ess because ising pe o mance equi emen s
and secu i y needs, along wi h a ailabili y needs, equi e ad ancemen . Mode n cloud ne wo ks di e om adi ional
da a cen e s because hey employ dynamic sys ems ha consis o i ualized in as uc u e as well as ansien se ices
ha ex end be ween a ious geog aphical egions ac oss p o ide s. Readiness o handle ine-g ained a ic con ol
es s on deployable ne wo k pla o ms, namely SDN, oge he wi h NFV and edge compu ing, which c ea e addi ional
ne wo k complexi y.
The mission o main aining e icien a ic low while a oiding conges ion and p o iding Quali y o Se ice oge he
wi h cybe h ea p o ec ion has become ex ensi ely di icul du ing his pe iod. T adi ional ne wo k managemen
me hods buil wi h heu is ics and se ou ing policies show inadequa e lexibili y when i comes o eal- ime
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 868-873
869
pe o mance enhancemen o anomaly esponse. The s a ic de ense app oaches p o e ine ec i e agains con empo a y
h ea s like applica ion-laye DDoS a acks alongside silen cloud enan la e al ac i i ies.
A i icial In elligence, h ough i s ein o cemen lea ning (RL), deep lea ning, and g aph analy ical sys ems, p esen s an
e ec i e al e na i e pe spec i e. These echnological solu ions allow ne wo ks o s udy eleme y in o ma ion,
es ablish a ic pa e ns, and o ecas conges ion poin s ollowed by ou e con ol au oma ion. Rein o cemen lea ning
agen s imp o e hei s a egies by di ec ly in e ac ing wi h he ne wo k o op imize pe o mance ac o s and achie e
eal- ime eedback on h oughpu la ency and packe losses. Th ough deep neu al ne wo k p ocessing o low da a along
wi h packe heade s, he sys em ob ains he capabili y o disco e hidden pa e ns ha eed in o a ack de ec ion.
The pape explo es how a i icial In elligence modi ies cloud ne wo king unc ions h ough dynamic a ic enginee ing
and QoS-awa e ou ing, as well as in elligen h ea de ec ion measu es. The s udy o ML algo i hms applied o eleme y
in o ma ion consis ing o Ne Flow logs ou ing upda es and ne wo k opology g aphs seeks o demons a e scalable,
in elligen ne wo king solu ion implemen a ion me hods. The analysis includes eal-wo ld case examples ha
demons a e ope a ional changes due o AI-enabled cloud ne wo ks, along wi h a discussion o implemen a ion and
e hical aspec s o sa e deploymen .
2. Challenges in In elligen Cloud Ne wo king
Cloud-na i e sys ems enable au oma ic wo kload managemen h ough scale-on-demand ope a ions, which p oduce
unp edic able and apidly changing a ic beha io s. T a ic dis ibu ion su e s immedia e majo changes hanks o
e en s such as au oscaling, mic ose ice eplica ion con en deli e y su ges, and egional ailo e s. OSPF and BGP,
oge he wi h adi ional ou ing algo i hms, expe ience slow eac ion imes, which leads o la e e ou ing a emp s
a e conges ion and delayed la ency occu . Se ice- o-se ice a ic pa e ns exis besides he clien -se e low, which
makes ou ing decisions mo e di icul o de e mine (Alhaida i, e al., 2021). Se ice meshes oge he wi h con aine ized
sys ems adjus hei a ic di ec ions because o a ailabili y zone condi ions and se e al se ice ou ing c i e ia,
including load-balancing ea u es and use iden i ica ion schemes. Models a e needed o dynamically lea n and
au onomously espond o modi ied ne wo k opologies as well as a ic demands and applica ion con ol objec i es in
a sys em lacking human in ol emen .
The con empo a y cloud en i onmen suppo s di e en applica ion ypes, which include bo h low-la ency c i ical
sys ems and high-bandwid h sys ems ha co e eal- ime communica ion, ideo s eaming, machine lea ning asks,
and IoT eleme y capabili ies. Di e en a ic ypes a e con ined o using he same physical and i ual ne wo k
connec ions, hus c ea ing complica ions o end- o-end pe o mance assu ance. T a ic bu s s and eme gency
conges ion issues canno be esol ed h ough ixed con igu a ions ha include a e limi a ions, queue weigh ules, and
oken bucke il e s o conges ion con ol. Limi ed bandwid h sa u a ion be ween mul iple se ices esul s in QoS
policy deg ada ion ha p oduces packe loss along wi h ji e e ec s and un esponsi e applica ions. Sha ed i ual
ne wo ks su e om majo pe o mance loss as enan s in e e e wi h one ano he . Se ice le el objec i es (SLOs) need
dynamic and p edic i e QoS en o cemen mechanisms ha con inuously adap o ne wo k s a e in o de o main ain
hem.
The achie emen o deep obse abili y emains d ama ically di icul wi hin so wa e-de ined cloud en i onmen s. The
combina ion o i ualized ne wo k unc ions (VNFs), o e lay unnels, and p og ammable da a planes p oduces
ne wo king abs ac ion ha hinde s eal- ime pe o mance acking as well as packe -le el p oblem de ec ion.
Pe o mance eleme y da a exis s in sepa a e pa s o he sys em as applica ion p oxies main ain esponse ime logs.
A he same ime, SDN con olle s keep ack o low ules, and hype iso s moni o bandwid h, bu he h ee domains
equi e ex ensi e wo k o connec hei eco ds. Epheme al se ices like con aine s ha s a and inish quickly
p oduce e alua ion di icul ies o he ope a o s. Ope a ions eams s uggle o moni o pe o mance and secu i y wi h
an in eg a ed de ailed unde s anding ha enables hem o ul ill hei du ies e ec i ely (Zhao, e al., 2019).
New secu i y isks eme ge because SDN sys ems a e bo h cen alized and p og ammable. A acke s can pene a e SDN
con olle s because hese ne wo k managemen cen e s se e as he ope a ional command cen e s whe e low
managemen esides. The a acke s use he in il a ion o place dange ous p o ocols in o he ne wo k. P oblems ha
eme ge om con olle e o s and API mal unc ions, oge he wi h inco ec low able con igu a ions, will sp ead
h oughou he ne wo k o a ec a ious se ice componen s and enan sys ems simul aneously. P og ammable ules
ha handle on -end a ic gene a e addi ional a ack oppo uni ies du ing DDoS assaul s because hey c ea e ou ing
ine iciencies and backend se ice o e load. La e al mo emen occu s h ough misused logical segmen a ion in mul i-
enan en i onmen s. Con empo a y secu i y s anda ds canno wa ch o e and s op hese up- o-da e dynamic h ea s
wi hin eal- ime sys em pa ame e s.
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 868-873
870
A majo i y o cloud ne wo ks depend on manual ou ing policies oge he wi h i ewall ules and QoS con igu a ions
ha equi e ex ensi e ime o modi y o dynamic changes. Adminis a i e s a needs o pe o m manual policy
adjus men s o new applica ion deploymen s and a ic- ela ed egional e en s since his p ocess akes oo long and
leads o human e o s wi hou scalable possibili ies. The cu en eac i e ne wo k s a us c ea es inadequa e
pe o mance and eliabili y le els h oughou he ne wo k. Ne wo ks canno adjus hemsel es owa ds be e
pe o mance me ics ia au oma ed eedback p ocesses when AI is absen om he analysis. Lacking au onomy es ic s
o ganiza ions om de eloping cloud in as uc u es wi h au onomous op imiza ion ea u es ha con o m o cu en
De Ops and CI/CD ope a ional models.
3. Solu ions in In elligen Cloud Ne wo king
Cloud en i onmen s ha ad ance in o dynamic p og ammable ne wo ks can be managed by A i icial In elligence (AI)
because i p o ides necessa y adap abili y and scalabili y wi h p edic i e powe o complex ne wo k beha io con ol.
This pa explains how ein o cemen lea ning and deep lea ning ope a e wi h g aph-based in e ence o eshape cloud
ne wo king ope a ions in a ic enginee ing ields and enac QoS p o ocols and cybe h ea ecogni ion sys ems.
The usage o Rein o cemen Lea ning (RL) p oduces an e ec i e amewo k ha op imizes a ic ou ing unde
condi ions o dynamic and unce ain en i onmen s. The ne wo k en i onmen ac s as he aining g ound o RL agen s
who na iga e con inuously while ga he ing pe o mance eedback, which ela es o la ency in addi ion o h oughpu
and packe loss me ics (Nanda, 2023). The agen s make op imal ou ing o load-balancing decisions by unning policies
lea ned h ough algo i hms Deep Q-Ne wo ks (DQN) P oximal Policy Op imiza ion (PPO) o Ac o -C i ic app oaches
based on changing ne wo k s a es and a ic condi ions.
SDN makes i possible o in eg a e RL agen s wi h he SDN con olle in o de o p ocess eal- ime ne wo k eleme y
ollowed by p ecise o wa ding decisions. An RL-based sys em applies au oma ed a ic di e sion o p e en link
conges ion and au oma ically adjus s Kube ne es clus e eas -wes low dis ibu ion. The implemen a ion o RL leads
o highe ne wo k link u iliza ion, and i sho ens he low comple ion pe iod, especially du ing hea y a ic pe iods.
When dealing wi h bo h ede a ed and mul i-cloud scena ios, hie a chical RL p o ides cen al domain con ol unde
independen local ope a ional condi ions. Th ough his app oach, cloud ne wo ks gain au onomous abili ies o adjus
ne wo k ope a ions in esponse o sudden changes in wo kload sys em ailu es and inc eased usage.
AI demons a es supe io pe o mance han s a ic mechanisms when i comes o p edic i e QoS op imiza ion. A i icial
in elligence models o bo h supe ised and unsupe ised ca ego ies analyze me ics om pas and cu en e en s o
p edic conges ion and manage bandwid h in ad ance h ough o ecas ing. The modeling o QoS beha io in cloud
backbones o da a cen e ab ics uses au o-encode s alongside ime-se ies o ecas ing models, including LSTMs as well
as hyb id decision ees.
AI sys ems acqui e knowledge abou wo kload beha io in a ious ne wo k condi ions, which allows hem o assign
low classes o au oma ed se ice speci ica ion implemen a ion. The ne wo k p o ides gua an eed bandwid h wi h
educed la ency o VoIP and AR/VR applica ions, which a e labeled as high p io i y while allowing ba ch jobs o ecei e
bes -e o ou es. The moni o ing o QoS iola ions be ween i ual ne wo k unc ions (VNFs) by AI enables a ic
ealloca ion and se ice mig a ion p ocedu es. The in eg a ion o a i icial In elligence exis s wi hin bo h a ic shape s
and conges ion con ol algo i hms o ce ain sys ems o dynamically adjus pe o mance h ough low pa e n
ecogni ion. The QoS en o cemen sys em wi h AI models exceeds adi ional ule en o cemen h ough con inuous
adjus men o main ain se ice quali y ac oss mul i- enan cloud pla o ms (La ah & Toke , 2019).
Ad anced h ea s such as DDoS a acks, la e al mo emen API abuse, and con ol-plane manipula ion a ge cloud
ne wo ks equen ly, and hese h ea s escape de ec ion by adi ional signa u e-based in usion de ec ion sys ems.
Schola ly deep lea ning sys ems analyze bo h seman ic pa e ns and beha io al indica ions o de ec ha d- o-de ec
o ms o a acks h oughou ne wo k laye s. Ne wo king da a de i ed om packe heade s unde go analysis by CNNs
oge he wi h LSTM ne wo ks and T ans o me s, which de ec no mal a ic pa e ns using in o ma ion collec ed om
SDN logs.
The aining o an LSTM model on low sequences allows i o iden i y h ee ypes o ne wo k anomalies, including po
scans and s eal hy command-and-con ol ac i i y, as well as apid session bu s s om bo ne s. T ans o me s
demons a e excellen pe o mance in unde s anding mul i- ield packe heade s o conduc deep packe analysis
wi hou manual ule c ea ion. AI sys ems in SDN ne wo ks use hei capabili ies o iden i y ogue ule inse ions as well
as disco e uno hodox con ol-plane inpu commands. P og ammable swi ches like hose based on P4 combine
e ec i ely wi h hese models o pe o m eal- ime olume ic a ack managemen a ne wo k edges and be o e a acks
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 868-873
871
each cen al compu e esou ces. Secu i y de ec ion sys ems can bene i om in en -based ne wo k policies because
hey enable au oma ic ac ions such as suspicious low qua an ine o edi ec hem o be e cloud secu i y esponse.
4. Case S udies and Examples
The ollowing pa illus a es p ac ical ins ances o a ic op imiza ion wi h ein o cemen lea ning while QoS
main enance and cybe de ense h ough AI deploymen in eal-wo ld cloud ne wo ks. The analyzed solu ions include
public cloud acili ies oge he wi h hype scale in as uc u e and so wa e-de ined solu ions ha ope a e wi hin
mul iple indus ial sec o s.
Mic oso Azu e ope a es as a wo ldwide cloud pla o m ha se es millions o use s by implemen ing ein o cemen
lea ning (RL) o i s WAN a ic con ol sys ems. The sys em named DeepCon u ilizes deep RL agen s o manage SD-
WAN ou ing dynamics ac oss Azu e's backbone ne wo k. The sys em uses agen s ha analyze link u iliza ion oge he
wi h pa h la ency measu emen s along wi h queue leng hs o execu e pa h-swi ching ope a ions. This me hod p o ided
quick accommoda ion o ansien link ailu es because i achie ed c i ical a ic e ou ing wi hou adi ional BGP
con e gence delays. The DeepCon sys em enhanced policy s a egies, which esul ed in imp o ed low comple ion
imes be ween 15 and 30 pe cen h ough i s in e - egion links. The API in eg a es wi hin an SDN con olle o Azu e
o conduc con inuous lea ning abou changing a ic pa e ns while p o ing he po en ial use o AI-based ou ing a
scale ac oss p oduc ion cloud ne wo ks.
The main Chinese AI and cloud se ices p o ide , Baidu, employs machine lea ning echnology o manage i s da a cen e
conges ion. Baidu de eloped an LSTM-based p edic ion model o o ecas conges ion h ough deep lea ning analysis o
wo kload signa u es and swi ch da a in i s managemen o unp edic able a ic om AI aining jobs, ideo se ices,
and mobile apps (Zhang e al., 2023). Bandwid h ese a ions, as well as pa h p io i iza ion, ake place in ad ance
because o he model's unc ionali y. Wi hin Baidu's AIOps s a egy s ands his adap i e QoS engine, which enables li e
a ic iden i ica ion, queue o ganiza ion, and speci ic o wa ding acco ding o AI-based p edic ions ins ead o ules.
The sys em de ec s abno mal a ic beha io using beha io al anomaly de ec ion models, which iden i y sudden SYN
packe su ges, non- ypical geog aphic pa e ns, and i egula p o ocol pa e ns ha di e om his o ical da ase s.
These analy ical models a e upda ed con inually h ough a ic da a p ocessing measu ed in he ange, which enables
he pla o m o de ec new DDoS me hods. An ex ensi e UDP ampli ica ion a ack was s opped wi hin a minu e when
en opy examina ion led o immedia e ac ion h ough AWS Global Accele a o . The implemen a ion o a i icial
in elligence sys ems igge s au oma ic esponses ha need no cus ome ac ion while imp o ing he speed o h ea
de ec ion ime o a minimum o olume ic dange s.
Google Cloud implemen s he machine-lea ning-based And omeda backbone o manage i s cloud ne wo king se ices
ha ope a e wi hin in e nal ne wo ks. Google u ilizes p edic i e ML models o ack he condi ions o i s ne wo k links
oge he wi h ongoing a ic pa e ns and se ice quali y me ics in i s wo ldwide in as uc u e (Mish a e al., 2019).
The ein o cemen lea ning componen in Bandwid h En o ce le s i ual ne wo k h oughpu alloca e changes
acco ding o cus ome esou ce usage pa e ns and se ice deploymen ankings. The sys em implemen s op imal
policies o conges ion-awa e shaping h ough which i can p o ide ai ness while main aining QoS du ing pe iods o
hea y con en ions. Google u ilizes clus e ing and deep lea ning echnology wi hin i s ne wo k anomaly de ec ion ools
o ecognize ou e leaks as well as hijacks along wi h miscon igu a ions. Google Cloud achie ed SLO-based high-
h oughpu ne wo king capabili y h ough i s in eg a ion o p og ammable in as uc u e and AI echnology o
BigQue y ope a ions and AI aining clus e needs.
5. E hical and Implemen a ion Conside a ions
In o ma ion echnology sys ems ha use a i icial In elligence now ope a e h ough machine-lea ning capabili ies
du ing ne wo k in eg a ion. These echnologies boos ope a ional e iciency and adap i eness as well as secu i y
measu es, bu hey in oduce i al p oblems abou ai ness alongside anspa ency s anda ds, ad e sa ial esilience,
and ope a ional sus ainabili y. The deploymen o in elligen ne wo king solu ions equi es p ope a en ion o
iden i ied isks o bo h esponsible deploymen and e ec i e ope a ion.
Ne wo k a ic da a ed o AI sys ems migh unin en ionally lea n imp ope and incomple e unde s anding o p e ious
ne wo k da a. T aining models based mos ly on No h Ame ican da a cen e wo kloads cause hem o pe o m poo ly
when p ocessing a ic pa e ns om Asian o A ican e i o ies (Yang, 2019). The use o aining da a consis ing
mainly o benign a ic cha ac e is ics may cause he model o become less igilan owa d newly occu ing o
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 868-873
872
in equen o ms o a acks. The ou come o such biased aining leads o inapp op ia e classi ica ion decisions a ec ing
pa icula g oups o enan s and p o ocols o geog aphic a eas di e en ly. When mul iple use s sha e a single cloud
en i onmen , pe o mance bias can igge unnecessa y se ice limi a ions o indi idual subsc ibe s. The p ope
assessmen o AI models includes con inuous ai ness moni o ing in di e en egions as well as speci ic applica ions,
oge he wi h di e se da ase aining and adhe ence o ad e sa ial obus ness es ing ac oss en i onmen s and
p o ocols.
Ne wo k ope a o s canno easily unde s and black-boxed ope a ions execu ed by ein o cemen lea ning agen s, deep
neu al ne wo ks, and ensemble classi ie s when hey pe o m ou ing changes, QoS adjus men s, and DDoS mi iga ion
p ocedu es. When o ganiza ions ail o main ain anspa ency abou hei ope a ion, his lowe s eliabili y and blocks
c i ical inciden esponse p o ocols. The app o al p ocess o au oma ed pa h e ou ing needs di ec explana ions om
he AI sys em ega ding i s ou e selec ion combined wi h in o ma ion abou he e alua ed ade-o s. Fundamen al
in e p e abili y p oblems in in elligen ne wo king sys ems can be esol ed wi h echniques ha deli e explana ions
h ough SHAP alues saliency maps and a en ion-based isualiza ions. The ools e eal how he AI eaches i s ou pu
decisions, which hen es ablishes us in au oma ed assis ance o ope a o s o make decisions.
P oduc ion cloud ne wo ks expe ience subs an ial ope a ional bu dens when hese ne wo ks deploy AI sys ems.
Regula e aining p ocedu es a e equi ed o models because hey need o adap o e ol ing a ic condi ions and
wo kload pa e ns oge he wi h secu i y h ea changes. MLOps pipelines need d i de ec ion oge he wi h ea u e
e olu ion and model e sion managemen as hei key ope a ional componen s. Cloud ne wo king eams cu en ly ace
p oblems because hey lack su icien employees who possess he equi ed skills ac oss all aspec s o da a enginee ing,
model aining, deploymen , and ollback. The imp ope execu ion o model go e nance leads o obsole e policies and
useless ale s, oge he wi h inco ec mi iga ion in e en ions. MLOps amewo ks o eal- ime ne wo k applica ions
equi e o ganiza ion in es men h ough au oma ed e aining sys ems combined wi h new model cana y es ing and
ools ha moni o model and ne wo k pe o mance dynamics.
Comme cial SDN pla o ms, as well as cloud AI se ices, combine hei in elligen ne wo king capabili ies in o APIs and
eleme y o ma s, which emain p op ie a y o cus ome s. The in eg a ion o hese ools p o es easy, bu hey c ea e
challenges when a emp ing model and policy ans e s h ough mul i-cloud o hyb id deploymen s. Pla o m models
can p o e di icul o audi since hei p e- ained elemen s lack anspa ency, and nobody has pe mission o access he
ained da a. O ganiza ions should implemen modula sys ems and adop open eleme y s anda ds oge he wi h
model o ma s o achie e po abili y when using in e ence engines. The o ganiza ion's abili y o hold in e nal con ol
o aining da a along wi h model a i ac s and e alua ion pipelines main ains independence and adap abili y o he
u u e needs o bo h in as uc u e and business ope a ions.
6. Conclusion
Cloud in as uc u e de elopmen owa d la ge scale and elas ic unc ionali y equi es in ense ne wo k a chi ec u al
adjus men s. Mode n cloud en i onmen s equi e ad anced p o ec ion measu es because s a ic ou ing, oge he wi h
ixed QoS policies and ule-based in usion de ec ion, canno adequa ely handle he complex na u e and secu i y h ea s
ha a ise in cu en cloud sys ems. A i icial In elligence (AI) has eme ged as an essen ial echnology o adap i e cloud
ne wo king because i p o ides au oma ic ne wo k con ol and eal- ime op imiza ion alongside adap i e de ense
capabili ies.
This analysis examines he p ima y ne wo k adminis a ion ba ie s aced by cloud sys ems due o unan icipa ed a ic
changes and ne wo k slowing, limi ed obse a ion capabili ies, g owing secu i y h ea s, and ixed policies ha ail o
adap . This pape e iews he applica ion o ein o cemen lea ning oge he wi h deep lea ning and g aph-based
models o imp o e a ic enginee ing QoS op imiza ion and h ea de ec ion unc ions. The implemen a ion o
ein o cemen lea ning p oduces au oma ed a ic con ol me hods, bu supe ised and unsupe ised models help
p edic conges ion and main enance quali y e ec s. Technology-based in usion de ec ion ools combine deep lea ning
algo i hms wi h mul i-modal da a analysis o p o ec p og ammable ne wo ks om complex ne wo k a acks.
Real-wo ld implemen a ions a Mic oso Azu e, AWS, Baidu, Google Cloud, and Cloud la e illus a e he ad anced s a e
and subs an ial bene i s achie ed h ough hese me hods ha enhance low comple ion imes and h ea p o ec ion
speeds oge he wi h Quali y-o -Se ice deli e y (Huang e al., 2020). The ad an ages o hese sys ems in ol e p ac ical
issues as well as mo al conce ns ega ding bo h model complexi y and algo i hm disc imina ion opposing a acks, as
well as he expanding wo kload o MLOps o eal- ime sys ems. The success ul implemen a ion o in elligen cloud
ne wo king solu ions needs bo h echnological p ecision and company p epa edness, as well as p ope compliance
models and hones managemen p ac ices.
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 868-873
873
In elligen cloud ne wo king sys ems will es ablish hemsel es as essen ial componen s equi ed o build a digi al
in as uc u e ha deli e s pe o mance alongside esilience and secu i y. AI echnology will con inue o de elop such
ha compu e ne wo ks will acqui e sel -go e nance capabili ies o sense eason and ac wi hou needing much human
in e ac ion. C i ical o ganiza ions ha app oach his ansi ion mind ully will achie e bo h expanded se ice speeds
and enhanced cus ome expe iences, as well as s onge de ensi e capabili ies du ing he cu en complex cloud
mo phing s age.
Re e ences
[1] Alhaida i, F., Almo i i, S. H., Al Ghamdi, M. A., Khan, M. A., Rehman, A., Abbas, S. ... & Rahman, A. U. (2021).
In elligen so wa e-de ined ne wo k o cogni i e ou ing op imiza ion using deep ex eme lea ning machine
app oach. Compu e s, Ma e ials & Con inua, 67(1), 1269-1285.
[2] Huang, Y., Cheng, Z., Zhou, Q., Xiang, Y., & Zhao, R. (2020). Da a mining algo i hm o cloud ne wo k in o ma ion
based on a i icial in elligence decision mechanism. IEEE Access, 8, 53394-53407.
[3] La ah, M., & Toke , L. (2019). A i icial in elligence enabled so wa e‐de ined ne wo king: a comp ehensi e
o e iew. IET ne wo ks, 8(2), 79-99.
[4] Mish a, D., Buyya, R., Mohapa a, P., & Pa naik, S. (2019). In elligen and cloud compu ing. P oceedings o ICICC, 1.
[5] Nanda, R. (2023). AI-Augmen ed So wa e-De ined Ne wo king (SDN) in Cloud En i onmen s. In e na ional
Jou nal o A i icial In elligence, Da a Science, and Machine Lea ning, 4(4), 1-9.
[6] Yang, S. (2019). Analysis o he eliabili y o compu e ne wo k by using in elligen cloud compu ing
me hod. In e na ional Jou nal o Compu e s and Applica ions, 41(4), 306-311.
[7] Zhang, L., Peng, J., Zheng, J., & Xiao, M. (2023). In elligen cloud-edge collabo a ions assis ed ene gy-e icien
powe con ol in he e ogeneous ne wo ks. IEEE T ansac ions on Wi eless Communica ions, 22(11), 7743-7755.
[8] Zhao, Y., Li, Y., Zhang, X., Geng, G., Zhang, W., & Sun, Y. (2019). A su ey o ne wo king applica ions applying he
so wa e de ined ne wo king concep based on machine lea ning. IEEE access, 7, 95397-95417.
