REAL-TIME ADAPTIVE MACHINE LEARNING FOR OPERATIONAL OPTIMIZATION ACROSS GLOBAL TRANSPORTATION, ENERGY, AND INDUSTRIAL INFRASTRUCTURE

Re iew o Applied Science and Technology
Volume 04, Issue 02 (2025)
Page No: 697 – 726
Doi: 10.63125/7a4h2916
697
Md A iqu Rahman Khan1;
Abs ac
This s udy in es iga es he ole o eal- ime adap i e machine lea ning (AML) in
op imizing ope a ions ac oss global anspo a ion, ene gy, g id, and indus ial
in as uc u es. The esea ch adop s a quan i a i e, c oss-sec ional design, es ing
he cen al hypo hesis ha AML implemen a ion signi ican ly imp o es sec o al
pe o mance ou comes compa ed o adi ional ule-based o s a ic op imiza ion
me hods. Fou speci ic hypo heses we e o mula ed: H1, AML imp o es
anspo a ion e iciency by educing conges ion and enhancing h oughpu ; H2,
AML inc eases ene gy o ecas accu acy by educing p edic ion e o s such as
mean absolu e pe cen age e o (MAPE); H3, AML s eng hens g id s abili y by
imp o ing equency and ol age egula ion; and H4, AML enhances indus ial
eliabili y h ough p edic i e main enance and down ime educ ion. Da a we e
d awn om seconda y sou ces, including case s udies, empi ical epo s, and
in e na ional deploymen s, and analyzed h ough desc ip i e s a is ics, co ela ion
es ing, collinea i y diagnos ics, and mul iple eg ession models. The indings
p o ided consis en and s a is ically signi ican suppo o all ou hypo heses. Fo
anspo a ion sys ems (H1), AML demons a ed a s ong posi i e e ec (β = .62, R²
= .39, p < .01), con i ming ea lie e idence om adap i e a ic con ol
deploymen s ha machine lea ning-d i en sys ems ou pe o m ixed- ime
scheduling. Fo ene gy sys ems (H2), AML signi ican ly educed o ecas ing e o s
(β = .55, R² = .30, p < .01), aligning wi h p io li e a u e on he supe io i y o ML-based
models o e con en ional s a is ical me hods. In e ms o g id s abili y (H3), AML
imp o ed ol age and equency egula ion (β = .58, R² = .34, p < .01), ein o cing
he a gumen ha adap i e o ecas ing and eal- ime con ol a e essen ial o
esilien ene gy sys ems. Indus ial sys ems (H4) exhibi ed he s onges associa ion,
wi h AML con ibu ing o p edic i e main enance accu acy and down ime
educ ion (β = .64, R² = .41, p < .01), ex ending p e ious indings ha indus ial
In e ne o Things (IIoT) applica ions a e pa icula ly esponsi e o adap i e
lea ning echniques. O e all, he esul s demons a e ha AML is a signi ican
p edic o o ope a ional op imiza ion ac oss all ou domains, wi h indus ial
eliabili y and anspo a ion e iciency showing he s onges gains. These indings
ad ance he li e a u e by mo ing beyond simula ion-based alida ions and
p o iding empi ical, c oss-sec o al e idence o AML’s ans o ma i e ole in
in as uc u e op imiza ion.
Keywo ds
Adap i e Machine Lea ning; Real-Time Op imiza ion; T anspo a ion Sys ems;
Ene gy In as uc u e; Indus ial Ope a ions.
REAL-TIME ADAPTIVE MACHINE LEARNING FOR
OPERATIONAL OPTIMIZATION ACROSS GLOBAL
TRANSPORTATION, ENERGY, AND INDUSTRIAL
INFRASTRUCTURE
1 MS in Managemen In o ma ion Sys em, Lama Uni e si y, Texas, USA;
Email: a [email p o ec ed]; mkhan35@lama .edu;
Ci a ion:
Khan, M. A. R. (2025). Real-
ime adap i e machine
lea ning o ope a ional
op imiza ion ac oss global
anspo a ion, ene gy, and
indus ial in as uc u e.
Re iew o Applied Science
and Technology, 4(2), 697 –
726.
h ps://doi.o g/10.63125/7a4h
2916
Recei ed:
June 09, 2025
Re ised:
July 10, 2025
Accep ed:
Augus 16, 2025
Published:
Sep embe 20, 2025
Copy igh :
© 2025 by he au ho . This
a icle is published unde he
license o Ame ican Schola ly
Publishing G oup Inc and is
a ailable o open access.
Re iew o Applied Science and Technology
Volume 04, Issue 02 (2025)
Page No: 697 – 726
Doi: 10.63125/7a4h2916
698
INTRODUCTION
Real- ime adap i e machine lea ning e e s o compu a ional sys ems ha can lea n con inuously,
upda e hemsel es dynamically, and espond ins an ly o s eaming da a while ope a ing in
luc ua ing en i onmen s (Wang e al., 2020). Such sys ems di e ge om classical o line machine
lea ning models by embedding online lea ning, eedback loops, and sel -modi ica ion mechanisms
o adjus pa ame e s and s a egies in si u. In p ac ice, hey combine me hodologies d awn om
ein o cemen lea ning, me a-lea ning, con inual lea ning, and s eaming analy ics o main ain
model ele ance and pe o mance as he en i onmen e ol es. The co e a ibu es o eal- ime
adap i e ML include con ex awa eness, inc emen al lea ning, low-la ency in e ence, and
obus ness o dis ibu ion shi (Ullah e al., 2020). T adi ional s a ic models, by con as , a e ained on
his o ical da ase s and hen deployed wi hou ongoing adap a ion; hey may de e io a e in
pe o mance as he da a dis ibu ion d i s o no el modes eme ge. The design o adap i e lea ning
machines mus balance esponsi eness wi h s abili y, a oiding o e i ing o momen a y noise o
ins abili ies. In enginee ing such sys ems, a chi ec s mus conside he compu a ional pipeline—da a
inges ion, p ep ocessing, inc emen al upda ing, model adap a ion—and he go e nance o
eedback loops ha p e en ca as ophic o ge ing o unaway adap a ion (Kong e al., 2020). The
no ion o “adap i e op imiza ion” in his con ex poin s o sys ems ha no only lea n bu ac i ely
op imize decisions in eal ime, closing he loop be ween lea ning and ope a ional con ol. A
companion concep is eal- ime ope a ional op imiza ion, which e e s o he dynamic adjus men
o con ol a iables o s a egies ( ou ing, dispa ch, powe alloca ion) in esponse o cu en sys em
s a e, unde he guidance o con inuously upda ing models. Toge he , “ eal- ime adap i e machine
lea ning o ope a ional op imiza ion” ames a class o in elligen sys ems ha ac , lea n, and
ecalib a e con inuously in mission-c i ical in as uc u e se ings.
Figu e 1: O e iew o Real-Time Adap i e Machine Lea ning o Ope a ional Op imiza ion
Global in as uc u e sys ems— anspo a ion ne wo ks, ene gy g ids, and indus ial p oduc ion
sys ems—a e among he mos complex enginee ed sys ems humans deploy. They o en span
mul iple geog aphies, egula o y egimes, empo al scales, and ope a ional modali ies (Danish &
Za o , 2022; Ramegowda & Mish a, 2021). T anspo a ion sys ems include oad, ail, shipping, ai , and
in e modal logis ics; ene gy in as uc u e includes gene a ion, ansmission, dis ibu ion, s o age, and
demand-side elemen s; and indus ial in as uc u e spans manu ac u ing, p ocess plan s, supply
chains, and main enance sys ems. Each domain by i sel p esen s o midable challenges: high
dimensionali y, he e ogenei y o subsys ems, s ochas ici y in demand, exogenous dis u bances
(wea he , acciden s, supply shocks), and s ong in e dependencies. When conside ing combined o
Re iew o Applied Science and Technology
Volume 04, Issue 02 (2025)
Page No: 697 – 726
Doi: 10.63125/7a4h2916
699
c oss-domain op imiza ion, he complexi y mul iplies, since decisions in one domain (e.g. ene gy
dispa ch) a ec cons ain s in ano he (e.g. anspo a ion o aw ma e ials)(Danish & Kam ul, 2022;
Lee & Rhee, 2021). T adi ional con ol and op imiza ion amewo ks, o en elying on s a ic models,
heu is ics, o pe iodic e-planning, equen ly all sho unde apidly changing condi ions o scale.
Many eal-wo ld dis up ions—wea he e en s, supply chain shocks, sudden demand su ges equi e
low-la ency adap a ion, which is beyond he capabili y o slow ba ch upda es (Deepa & Thillaia asu,
2024; Jahid, 2022a). The in e na ional signi icance lies in he ac ha in as uc u e unde pins mode n
economies, global supply chains, and socie al wel a e: ailu es o ine iciencies in anspo a ion,
ene gy, o indus ial sys ems cascade ac oss bo de s and sec o s. Hence, imp o emen s in hei
ope a ional e iciency and esilience di ec ly enhance global sus ainabili y, secu i y, and economic
compe i i eness. In his landscape, eal- ime adap i e ML o e s a pa h owa d b idging high-le el
decision-making wi h ine-g ained esponsi eness ac oss di e se geog aphies and scales (Jahid,
2022b; Yao e al., 2021).
T anspo a ion sys ems ha e been among he ea lies and mos isible bene icia ies o eal- ime
adap i e machine lea ning. In in elligen anspo a ion sys ems (ITS), ML models ha e been used o
p edic conges ion, de e mine signal imings, op imize ou ing, and manage a ic lows (A i u &
Noo , 2022; Rebollo e al., 2001). Fo example, adap i e a ic signal con ol amewo ks such as
SURTRAC dynamically op imize signal iming in eal- ime, yielding a el ime educ ions o ~25% and
wai - ime educ ions o ~40% in pilo deploymen s (see Scalable U ban T a ic Con ol). In logis ics
and eigh , eal- ime ou e op imiza ion sys ems combine LSTM-based a ic o ecas ing wi h
ein o cemen lea ning o adjus deli e y pa hs on he ly (Hasan e al., 2022; Yao e al., 2021). These
sys ems inges GPS da a, wea he eeds, a ic senso s, and lee s a us o p opose dynamic e ou ing
(Henesey e al., 2006; Redwanul & Za o , 2022). In mul imodal logis ics, deep ein o cemen lea ning
has been used o ou e adjus men and anomaly de ec ion ac oss bo de s. In oad- anspo
co ido s, neu al ne wo k–based lea ning has helped op imize long-dis ance ou ing (e.g. Dakhla–
Pa is) unde sa e y, cos , and ime cons ain s. Q-lea ning and a ian s ha e been adap ed o
dynamic ehicle ou ing and a eling salesman– ype p oblems unde eal- ime cons ain s (Rezaul
& Mesbaul, 2022; S. Wang e al., 2020). Recen e iews o ML in eigh anspo a ion highligh i s u ili y
in a i al ime es ima ion, demand o ecas ing, ehicle ou ing, a ic p edic ion, and anomaly
de ec ion (Jiang e al., 2020; Hasan, 2022).
Figu e 2: Real-Time Adap i e Machine Lea ning in Di e en sec o
In he ene gy domain, eal- ime adap i e machine lea ning has been le e aged o add ess g id
a iabili y, demand o ecas ing, and dynamic powe alloca ion. The ansi ion o enewable
Re iew o Applied Science and Technology
Volume 04, Issue 02 (2025)
Page No: 697 – 726
Doi: 10.63125/7a4h2916
700
gene a ion in oduces in e mi en supply, which demands ine-g ained, esponsi e con ol o
main ain s abili y. ML me hods ha e become in eg al o mode n sma g ids, mic og ids, and
demand-side managemen (Abdelsalam e al., 2020; Ta ek, 2022). One case is he ORA-DL
amewo k, which in eg a es deep neu al ne wo ks, ein o cemen lea ning, and IoT o alloca e
esou ces, o ecas demand, and educe was age in eal ime—yielding ~93.38% p edic ion
accu acy, 96.25% g id s abili y, and 22.96% lowe ope a ing cos ela i e o benchma ks (Kam ul &
Oma , 2022; Ullah e al., 2020). Hyb id ML + op imiza ion amewo ks o demand-side managemen
ha e also been p oposed, combining p edic i e models and cons ained op imiza ion o indus ial-
scale sys ems. In hei e iew, (Xin e al., 2018) documen he inc easing applica ion o ML ac oss
gene a ion scheduling, demand o ecas ing, ene gy s o age, aul de ec ion, and g id esilience
asks. Some s udies adop ede a ed lea ning combined wi h digi al wins o manage he e ogenei y
and p i acy ac oss dis ibu ed g id nodes. The o e a ching esul is ha eal- ime adap i e ML helps
ene gy sys ems adap con inuously o luc ua ion in demand, gene a ion, and ne wo k opology,
imp o ing e iciency, educing losses, and enhancing esilience (Kam ul & Ta ek, 2022; Wang e al.,
2020).
The objec i e o his s udy is o conduc a quan i a i e analysis o eal- ime adap i e machine
lea ning o ope a ional op imiza ion ac oss global anspo a ion, ene gy, and indus ial
in as uc u e, emphasizing measu able imp o emen s in e iciency, esilience, and cos educ ion.
By applying a da a-d i en app oach, he esea ch seeks o e alua e pe o mance me ics such as
educed ansi delays, lowe ed ene gy consump ion, minimized down ime, and enhanced
h oughpu in indus ial p ocesses. Quan i a i e analysis se es as he ounda ion o isola ing he
angible impac o adap i e models compa ed o s a ic sys ems, highligh ing nume ical di e ences
in p edic i e accu acy, op imiza ion speed, and sys em eliabili y. In anspo a ion, he s udy aims
o quan i y gains in a el ime educ ion, lee u iliza ion e iciency, and emissions con ol h ough
dynamic ou e op imiza ion and adap i e a ic managemen . In ene gy in as uc u e, he goal is
o measu e he ex en o which eal- ime lea ning con ibu es o g id s abili y, enewable ene gy
in eg a ion, and demand- esponse accu acy, exp essed h ough key pe o mance indica o s such
as pe cen age educ ions in peak load and ope a ing cos s. Fo indus ial in as uc u e, he
quan i a i e objec i es include e alua ing p edic i e main enance accu acy, educ ion in
unplanned machine ailu es, and imp o emen s in p oduc ion line e iciency measu ed agains
baseline me ics. This ocus on quan i iable ou comes ensu es ha he analysis mo es beyond
heo e ical claims o deli e conc e e e idence o he scalabili y and ope a ional alue o adap i e
lea ning. Ano he laye o he objec i e is o compa e pe o mance ac oss egions and indus ies,
p o iding a global pe spec i e ha accoun s o a iabili y in sys em ma u i y, da a a ailabili y, and
ope a ional complexi y. By s uc u ing he esea ch a ound measu able benchma ks, he s udy aims
o ansla e he abs ac p omise o eal- ime adap i e machine lea ning in o conc e e nume ical
insigh s ha can guide decision-make s, alida e in es men s, and demons a e he ans o ma i e
ole o con inuous adap a ion in mode n in as uc u e op imiza ion.
LITERATURE REVIEW
The s udy o eal- ime adap i e machine lea ning o ope a ional op imiza ion ac oss anspo a ion,
ene gy, and indus ial in as uc u e has a ac ed g owing a en ion as o ganiza ions wo ldwide
g apple wi h he challenges o e iciency, esilience, and sus ainabili y in la ge-scale sys ems. A
li e a u e e iew in his a ea equi es si ua ing he discussion wi hin h ee o e lapping domains: he
heo e ical ounda ions o adap i e lea ning, i s sec o al applica ions, and he c oss-domain
in eg a ion challenges ha accompany global in as uc u e op imiza ion. Exis ing schola ship
e lec s di e se me hodological app oaches, anging om algo i hmic inno a ions in ein o cemen
lea ning and con inual lea ning, o empi ical s udies measu ing sys em-le el imp o emen s in
logis ics, g id managemen , and indus ial p oduc ion. The e iew also d aws upon in e disciplina y
sou ces, combining pe spec i es om enginee ing, ope a ions esea ch, in o ma ion sys ems, and
applied compu e science. While p io s udies es ablish he echnical easibili y and ope a ional
bene i s o adap i e lea ning, hey also unde sco e pe sis en challenges in scalabili y,
in e ope abili y, and sa e y-c i ical deploymen s. This sec ion sys ema ically e iews key s ands o he
li e a u e o p o ide cla i y on concep ual de ini ions, algo i hmic s a egies, empi ical applica ions
ac oss domains, and he compa a i e ad an ages and limi a ions epo ed in di e en con ex s. In
doing so, i iden i ies pa e ns and gaps ha shape he cu en unde s anding o adap i e machine
lea ning in in as uc u e op imiza ion and lays he ounda ion o a ocused quan i a i e analysis..
Re iew o Applied Science and Technology
Volume 04, Issue 02 (2025)
Page No: 697 – 726
Doi: 10.63125/7a4h2916
701
Real-Time Adap i e Machine Lea ning
The schola ly ounda ion o eal- ime adap i e machine lea ning is ancho ed in he e olu ion o
online lea ning, ein o cemen lea ning, and dynamic con ol sys ems. Ea ly s udies in con ol heo y
emphasized he need o sys ems ha could adjus o changing s a es and unce ain ies, which la e
in o med machine lea ning app oaches capable o con inual sel -adjus men (Abdelsalam e al.,
2020). In adap i e con ex s, models a e dis inguished om s a ic coun e pa s by hei abili y o
inc emen ally inco po a e new da a and adjus decision bounda ies wi hou e aining om sc a ch.
This p ope y is pa icula ly impo an o handling dis ibu ion d i , a ecu ing p oblem in non-
s a iona y en i onmen s whe e da a dis ibu ions e ol e o e ime. Adap i e ML is also cha ac e ized
by he s abili y–plas ici y balance, a dilemma ha conce ns main aining p io knowledge while
emaining esponsi e o no el in o ma ion (Li e al., 2018; Mubashi & Abdul, 2022). Li e a u e has
emphasized he ole o con inual lea ning as a mechanism o mi iga e ca as ophic o ge ing and
ensu e long- e m model iabili y. Rein o cemen lea ning, in pa icula , has been ad anced as a
ounda ional pa adigm o adap i e sys ems because o i s capaci y o upda e policies based on
en i onmen al eedback in eal ime. Theo e ical con ibu ions also highligh he impo ance o
la ency educ ion and obus ness in mission-c i ical deploymen s, whe e eal- ime op imiza ion
di ec ly impac s ope a ional sa e y and e iciency. Mo e ecen e iews expand on hyb id
app oaches, which in eg a e op imiza ion me hods wi h adap i e ML o enhance bo h
in e p e abili y and pe o mance (Muhammad & Kam ul, 2022; Zhao e al., 2018). Collec i ely, his
body o wo k es ablishes he concep ual g ounding o eal- ime adap i e machine lea ning as a
pa adigm si ua ed a he in e sec ion o dynamic sys ems heo y, compu a ional in elligence, and
applied op imiza ion.
Figu e 3: Real-Time Adap i e Machine Lea ning
Sou ce: Wu, Rincon and Ch is o ides. (2019)
The li e a u e on algo i hmic inno a ions in eal- ime adap i e machine lea ning demons a es apid
p og ess in ein o cemen lea ning, con inual lea ning, and ede a ed lea ning a chi ec u es.
Rein o cemen lea ning (RL) me hods, such as Q-lea ning and deep RL, a e equen ly applied o
en i onmen s whe e decisions mus e ol e dynamically wi h unce ain eedback. Mul i-agen RL has
been in es iga ed o dis ibu ed in as uc u e con ol, whe e mul iple au onomous en i ies
collabo a e unde adap i e policies. Ano he signi ican de elopmen is con inual lea ning, which
add esses ca as ophic o ge ing by in oducing eplay mechanisms, egula iza ion-based
s a egies, and a chi ec u al modula i y o p ese e p io knowledge (Reduanul & Mohammad
Shoeb, 2022; Wang e al., 2020). Online lea ning me hods ex end his ajec o y by allowing
inc emen al upda es o model weigh s as da a s eams a i e, enabling nea -ins an aneous
adap a ion in ope a ional se ings (Noo & Momena, 2022; Yao e al., 2021). Fede a ed lea ning has

Re iew o Applied Science and Technology
Volume 04, Issue 02 (2025)
Page No: 697 – 726
Doi: 10.63125/7a4h2916
702
eme ged as pa icula ly ele an in in as uc u e con ex s, since i pe mi s decen alized model
aining ac oss dis ibu ed nodes while p ese ing da a p i acy and educing communica ion
bo lenecks. In addi ion, hyb id amewo ks in eg a ing model p edic i e con ol wi h adap i e
lea ning ha e been es ed in cybe -physical sys ems, demons a ing enhanced obus ness and
in e p e abili y compa ed o s andalone ML app oaches. Ad ances in ans e lea ning u he
enable c oss-domain adap abili y, allowing models ained in one in as uc u e domain o be
ecalib a ed e ec i ely o ano he (A un e al., 2024; Danish, 2023). Taken oge he , hese
algo i hmic con ibu ions expand he ope a ional capaci y o adap i e ML, o e ing obus , scalable,
and con ex -awa e solu ions ha can be deployed in en i onmen s wi h high le els o a iabili y and
complexi y.
T anspo a ion esea ch has become one o he mos isible domains o eal- ime adap i e machine
lea ning, wi h s udies add essing a ic con ol, logis ics op imiza ion, and mul imodal in eg a ion.
Adap i e a ic signal con ol sys ems ep esen a ma u e line o inqui y, wi h empi ical s udies
showing ha ein o cemen lea ning–d i en adap i e signals educe conges ion and a el imes
signi ican ly compa ed o ixed- ime models. Fo ins ance, SURTRAC, an RL-based a ic signal
con ol sys em, demons a ed educ ions o 25% in a el ime and 40% in wai ing ime in u ban ials.
Logis ics applica ions emphasize dynamic ou e op imiza ion, whe e deep lea ning and RL
amewo ks imp o e deli e y e iciency unde luc ua ing demand and a ic condi ions. In
mul imodal con ex s, adap i e ML has been applied o op imize in e ac ions be ween oad, ail, and
shipping ne wo ks, yielding e iciency gains in eigh anspo and supply chain esponsi eness
(Giannocca o & Pon andol o, 2002; Hasan e al., 2023). P edic i e models o a i al imes based on
s eaming GPS and a ic da a u he illus a e how online lea ning app oaches enhance eliabili y
in public anspo . Mo e ecen applica ions in eg a e adap i e anomaly de ec ion wi h p edic i e
logis ics o handle dis up ions in c oss-bo de eigh sys ems. S udies o ide-sha ing sys ems also
highligh he u ili y o adap i e ML in eal- ime dispa ching and demand alloca ion, whe e
ein o cemen lea ning amewo ks ou pe o m heu is ic me hods. Collec i ely, hese con ibu ions
show ha anspo a ion in as uc u es bene i signi ican ly om adap i e ML, wi h quan i a i e
e idence o educed delays, op imized lee u iliza ion, and imp o ed se ice eliabili y ac oss
di e se in e na ional con ex s (Hossain e al., 2023). Applica ions o eal- ime adap i e machine
lea ning in ene gy and indus ial domains unde sco e i s ole in s abilizing g ids, enhancing e iciency,
and educing ope a ional isks. In ene gy sys ems, adap i e ML has been cen al o demand
o ecas ing, whe e models ha upda e con inuously ou pe o m s a ic p edic o s in cap u ing load
luc ua ions. Sma g id s udies highligh RL-based con olle s o demand esponse and dis ibu ed
gene a ion, showing imp o emen s in cos e iciency and s abili y (Hosein & Hosein, 2017).
Renewable in eg a ion, pa icula ly o wind and sola , has been enhanced h ough online lea ning
amewo ks capable o adjus ing p edic ions unde a iable me eo ological condi ions. In indus ial
con ex s, p edic i e main enance is a dominan applica ion, whe e adap i e ML models analyze
senso da a o an icipa e equipmen ailu es and educe unplanned down ime. Adap i e
op imiza ion also plays a ole in p ocess con ol, wi h hyb id ML–MPC amewo ks imp o ing
h oughpu and quali y in manu ac u ing en i onmen s. Indus ial In e ne o Things (IIoT) esea ch
demons a es how ede a ed lea ning can suppo decen alized op imiza ion in ac o ies while
sa egua ding sensi i e da a(Raza i-Fa e al., 2019). Supply chain applica ions emphasize adap i e
o ecas ing o dynamic esou ce alloca ion, enhancing esponsi eness o demand shocks and
anspo a ion delays. S udies consis en ly highligh measu able bene i s, including cos educ ions,
imp o ed eliabili y, and e iciency gains, posi ioning adap i e ML as a s a egic enable in ene gy
and indus ial in as uc u es globally.
Co e Theo e ical Cons uc s
Theo e ical discou se on adap i e machine lea ning emphasizes he impo ance o balancing
s abili y and plas ici y in eal- ime sys ems. The s abili y–plas ici y dilemma, i s desc ibed in cogni i e
neu oscience, e e s o he ension be ween e aining p io knowledge (s abili y) and in eg a ing new
in o ma ion (plas ici y) wi hou ca as ophic o ge ing (Li e al., 2019; Hossain e al., 2023). This
challenge has been ex ensi ely s udied in machine lea ning, pa icula ly wi hin con inual lea ning
amewo ks. Algo i hms such as elas ic weigh consolida ion and memo y-based eplay me hods
a emp o p ese e s abili y while allowing o adap a ion. In online lea ning, models mus
inco po a e new da a s eams inc emen ally, wi h heo e ical analyses highligh ing ade-o s
be ween con e gence speed and model obus ness. Rein o cemen lea ning p o ides addi ional
Re iew o Applied Science and Technology
Volume 04, Issue 02 (2025)
Page No: 697 – 726
Doi: 10.63125/7a4h2916
703
g ounding, as policy upda es e lec ongoing plas ici y, while alue unc ion s abiliza ion ancho s
long- e m pe o mance. Empi ical e alua ions o adap i e algo i hms ac oss non-s a iona y
en i onmen s unde sco e he agili y o s abili y when aced wi h ab up dis ibu ion shi s (Uddin &
Ash a , 2023; Ullah e al., 2020). Complemen a y app oaches such as me a-lea ning u he highligh
he capaci y o sys ems o ecalib a e plas ici y h esholds dynamically, enabling as e adap a ion
ac oss asks ((Momena & Hasan, 2023; Wang e al., 2020). Collec i ely, he li e a u e posi ions he
s abili y–plas ici y ade-o as a cen al heo e ical cons uc ha guides bo h he design and
e alua ion o adap i e sys ems.
Figu e 4: Theo e ical F amewo k o his s udy
A second key cons uc conce ns he handling o non-s a iona y da a dis ibu ions, o en desc ibed
as dis ibu ion d i , which di ec ly impac s model alidi y in eal- ime se ings. S udies ca ego ize d i
in o g adual, ab up , and ecu ing pa e ns, each p esen ing dis inc challenges o adap i e
sys ems. D i adap a ion me hods include windowing s a egies, ensemble app oaches, and
p obabilis ic de ec ion mechanisms ha lag dis ibu ional changes. Fo example, adap i e andom
o es s ha e been p oposed o main ain p edic i e accu acy in e ol ing da a s eams by
inc emen ally upda ing ee ensembles (Mubashi & Jahid, 2023; Tools e al., 2018). Neu al ne wo ks
also exhibi imp o ed esilience when combined wi h d i de ec o s ha selec i ely igge e aining.
Real- ime ene gy demand o ecas ing s udies show how d i can unde mine s a ic models,
ein o cing he impo ance o con inuous ecalib a ion. Simila ly, anspo a ion applica ions highligh
ab up d i s caused by dis up ions such as acciden s o wea he e en s, necessi a ing models ha
adap wi hin seconds(Ganesh e al., 2024; Sanjai e al., 2023). Theo e ical wo k on concep d i
u he emphasizes i s ine i abili y in dynamic en i onmen s, sugges ing ha adap abili y mus be a
co e design p inciple a he han an auxilia y unc ion. Online Bayesian upda ing amewo ks also
demons a e s ong heo e ical g ounding o handling unce ain y in eal- ime adap a ion. Ac oss
di e se applica ions, he li e a u e con e ges on he ecogni ion ha d i - esilien lea ning is
undamen al o sus ained ope a ional op imiza ion in dynamic in as uc u es.
Rein o cemen Lea ning in Dynamic En i onmen s
Rein o cemen lea ning (RL) o malizes sequen ial decision making unde unce ain y h ough
Ma ko decision p ocesses (MDPs), whe e agen s lea n policies mapping s a es o ac ions o
maximize cumula i e e u n (Cheung e al., 2002; Ak e e al., 2023). Ea ly ounda ions es ablished
alue-based lea ning and empo al-di e ence me hods, including Q-lea ning, which con e ges
Re iew o Applied Science and Technology
Volume 04, Issue 02 (2025)
Page No: 697 – 726
Doi: 10.63125/7a4h2916
704
unde ce ain condi ions in abula se ings. Func ion app oxima ion ex ended hese ideas o high-
dimensional p oblems bu in oduced ins abili y, mo i a ing algo i hmic designs ha ca e ully
manage boo s apping, o -policy lea ning, and non-s a iona y a ge s. Deep Q-Ne wo ks (DQN)
pai ed neu al unc ion app oxima o s wi h expe ience eplay and a ge ne wo ks o s abilize alue
lea ning om aw pixels, demons a ing obus con ol in isually ich, apidly changing en i onmen s.
Pa allel ad ances in policy sea ch led o policy-g adien me hods wi h con e gence gua an ees
unde mild assump ions and p ac ical a iance- educ ion echniques (Hosein & Hosein, 2017). T us
Region Policy Op imiza ion (TRPO) and P oximal Policy Op imiza ion (PPO) cons ained policy
upda es o p ese e mono onic imp o emen and empi ical s abili y unde dynamic condi ions
(Danish & Za o , 2024; Giannocca o & Pon andol o, 2002). Fo con inuous con ol, de e minis ic
policy g adien s and ac o –c i ic a ian s o e ed e icien lea ning in high-dimensional ac ion spaces
ypical o dynamic obo ic and indus ial se ings. So Ac o –C i ic (SAC) in oduced en opy-
egula ized objec i es ha encou age obus , di e se beha io s and s ong sample e iciency unde
shi ing dynamics. Dis ibu ional RL e amed alue lea ning o e e u n dis ibu ions, yielding be e
isk sensi i i y and empi ical pe o mance in changing ewa d landscapes. In eg a i e baselines such
as Rainbow combined p io i ized eplay, mul i-s ep e u ns, and dis ibu ional es ima es, illus a ing
cumula i e bene i s o s abili y-o ien ed componen s o dynamic en i onmen s (A un e al., 2024;
Jahid, 2024a). Collec i ely, hese o mula ions and algo i hms g ound RL’s capaci y o adap o
e ol ing s a e–ac ion con ingencies while main aining lea ning s abili y in complex se ings.
Figu e 5: Rein o cemen Lea ning in Dynamic En i onmen s
Dynamic en i onmen s expose agen s o shi ing ansi ion dynamics and ewa d s uc u es,
in ensi ying he explo a ion–exploi a ion dilemma and he need o sample-e icien lea ning (Jahid,
2024b; Tools e al., 2018). Coun -based and pseudo-coun explo a ion encou age isi s o no el
s a es, imp o ing adap abili y when en i onmen s a is ics change. Boo s apped ensembles
app oxima e pos e io unce ain y o d i e deep explo a ion and as e eco e y om non-
s a iona i y. In insic-mo i a ion s a egies— a ia ional in o ma ion gain, p edic ion-e o –based
cu iosi y, and empowe men —sus ain explo a o y beha io in spa se- ewa d o ab up ly changing
se ings. O -policy ac o –c i ic me hods imp o ed da a euse bu aced o e es ima ion and
di e gence isks; Twin Delayed DDPG (TD3) mi iga ed hese ia clipped double c i ics and a ge
policy smoo hing in con inuous con ol. Ba ch-cons ained and conse a i e o -policy algo i hms
u he s abilized lea ning om ini e bu e s, which is common when in e ac ion mus emain
bounded unde ope a ional cons ain s. Model-based RL (MBRL) enhances sample e iciency by
lea ning en i onmen dynamics and planning wi h imagined ollou s; PILCO achie ed s ong da a
e iciency wi h p obabilis ic dynamics, while mode n neu al ensembles imp o ed unce ain y
quan i ica ion o obus con ol unde changing dynamics (Lee & Rhee, 2021; Hasan, 2024). Sho -
Re iew o Applied Science and Technology
Volume 04, Issue 02 (2025)
Page No: 697 – 726
Doi: 10.63125/7a4h2916
705
ho izon model-based policy op imiza ion educed model bias by limi ing ollou leng h and blending
model- ee a ge s. Wo ld-model app oaches demons a ed ha compac la en dynamics enable
apid policy adap a ion when obse a ions shi . Toge he , hese s ands show how explici
unce ain y handling, p incipled explo a ion, and lea ned models con ibu e o esilien pe o mance
and apid e-op imiza ion when en i onmen al s a is ics a y.
Con inual and Inc emen al Lea ning
A cen al heo e ical and p ac ical challenge in con inual lea ning is ca as ophic o ge ing, whe e
models ained sequen ially on new da a apidly o e w i e knowledge om p e ious asks. Se e al
algo i hmic amilies ha e eme ged o mi iga e his e ec . Regula iza ion-based s a egies cons ain
upda es so ha weigh s c i ical o ea lie asks a e minimally al e ed. A canonical example is Elas ic
Weigh Consolida ion (EWC), which es ima es pa ame e impo ance h ough he Fishe In o ma ion
Ma ix and imposes a quad a ic penal y o de ia ing om p e iously op imized pa ame e s (Jahid,
2025a; Li e al., 2019). Building on his idea, Synap ic In elligence (SI) compu es an impo ance
measu e by accumula ing con ibu ion o loss educ ion du ing aining, allowing online es ima ion
wi hou equi ing ask bounda ies (Jahid, 2025b; Ramegowda & Mish a, 2021). Ano he ex ension,
Memo y Awa e Synapses (MAS), es ima es impo ance by measu ing he sensi i i y o ou pu s o
weigh pe u ba ions, pe mi ing use in ask- ee scena ios. Complemen ing egula iza ion,
knowledge dis illa ion amewo ks such as Lea ning wi hou Fo ge ing (LwF) p ese e he unc ional
beha io o olde models by aligning so p edic ions o he cu en model wi h hose o ea lie
e sions. In pa allel, eplay-based app oaches mi iga e o ge ing by ein oducing p e ious da a o
app oxima ions he eo . iCaRL, o ins ance, main ains exempla s and le e ages nea es -mean
classi ica ion o s abilize ecogni ion unde class-inc emen al condi ions. G adien Episodic Memo y
(GEM) adds cons ain s o op imiza ion so ha new g adien s do no ha m pe o mance on s o ed
exempla s, while A-GEM imp o es e iciency by p ojec ing upda es on o a single g adien e e ence.
Gene a i e eplay ep esen s ano he pa hway, as in Deep Gene a i e Replay (DGR), whe e a
gene a o p oduces syn he ic samples om olde asks o ehea se alongside new da a. Toge he ,
hese algo i hms e eal ha ca as ophic o ge ing can be a enua ed by s a egically p ese ing
knowledge ei he h ough cons ained op imiza ion, eplay, o a chi ec u al isola ion while
main aining plas ici y o acqui ing new pa e ns. Elas ic Weigh Consolida ion (EWC) Loss Func ion:
Inc emen al lea ning me hods emphasize con inuous adap a ion in li e en i onmen s, whe e models
mus upda e e icien ly wi h new da a s eams while minimizing eg ession on pas knowledge. Online
op imiza ion algo i hms such as Online G adien Descen (OGD) and Follow-The-Regula ized-Leade
(FTRL) p o ide he heo e ical unde pinnings o sequen ial upda es, adjus ing model pa ame e s
wi h each new obse a ion unde bounded eg e gua an ees. In p ac ical deep lea ning, op imize s
such as Adam and RMSP op a e adap ed o s eaming con ex s by uning lea ning a es and
inco po a ing exponen ial decay o s abili y. Inc emen al Bayesian amewo ks, including Kalman
il e s and online Expec a ion-Maximiza ion, u he o malize con inual pa ame e upda ing wi h
p incipled unce ain y quan i ica ion, making hem especially sui able o senso - ich o sa e y-c i ical
applica ions. In deployed deep neu al models, ligh weigh adap a ion s a egies ha e p o en
e ec i e: Adap e modules and LoRA in oduce small ainable componen s o low- ank
decomposi ions in o ozen ne wo ks, allowing apid upda es wi hou ca as ophic eg ession on p io
asks. S eaming en i onmen s o en bene i om memo y bu e s, whe e Expe ience Replay (ER) wi h
ese oi sampling main ains ep esen a i e samples, and balanced ine- uning s a egies use hese
exempla s o a oid bias owa d new classes. D i de ec ion algo i hms such as ADWIN and Page-
Hinkley es s iden i y shi s in da a dis ibu ions, igge ing adap i e eweigh ing o bu e e eshes. Fo
policy-based sys ems, o -policy e alua ion echniques such as doubly obus es ima ion (Raza i-Fa
e al., 2019) allow sa e alida ion o inc emen al upda es be o e ull deploymen . Collec i ely, hese
me hods illus a e how inc emen al upda ing is ope a ionalized in p ac ice: combining e icien
online op imiza ion, unce ain y-awa e es ima ion, ligh weigh modula adap a ion, and d i
de ec ion o main ain model accu acy and eliabili y unde con inuously changing condi ions.
Re iew o Applied Science and Technology
Volume 04, Issue 02 (2025)
Page No: 697 – 726
Doi: 10.63125/7a4h2916
712
logis ics co ido s in Eu ope, such as he T ans-Eu opean T anspo Ne wo k (TEN-T), inco po a e
adap i e ML in a ic managemen and mul imodal in eg a ion, ensu ing smoo he low o goods
and educing bo de conges ion. In Asia, collabo a ions unde he ASEAN Sma Ci ies Ne wo k b ing
oge he ci ies ac oss Sou heas Asia o sha e digi al in as uc u e, da a go e nance s a egies, and
AI-enabled solu ions o mobili y and esou ce e iciency. In e na ional ene gy and sus ainabili y
amewo ks, such as Mission Inno a ion and IEA sma g id ini ia i es, u he emphasize he
impo ance o adap i e ML o ansna ional enewable in eg a ion and demand managemen (Li
e al., 2019). No h Ame ica has ad anced c oss-bo de collabo a ions, such as U.S.–Canada
ini ia i es in sma g ids and cybe secu i y o c i ical in as uc u es, whe e ede a ed lea ning and
ML-based anomaly de ec ion allow secu e ye dis ibu ed op imiza ion. Compa a i e e iews
highligh ha scalabili y challenges o en a ise om di e ing egula o y s uc u es, da a p i acy laws,
and in as uc u e ma u i y, equi ing ha monized go e nance and in e ope able s anda ds.
None heless, in e na ional collabo a ions p o ide obus es beds o ML-d i en in as uc u es,
demons a ing how adap i e sys ems can anscend local deploymen s o o m esilien , global
sma in as uc u e ne wo ks. The li e a u e unde sco es ha scalabili y eme ges mos e ec i ely
when ci ies and na ions in eg a e ML solu ions no in isola ion bu wi hin collabo a i e, c oss-bo de
amewo ks designed o sha e knowledge, mi iga e isks, and s anda dize digi al in as uc u e
de elopmen .
Resea ch Gaps
A ecu ing gap in he li e a u e on adap i e machine lea ning in in as uc u e sys ems is he
absence o longi udinal s udies ha e alua e pe o mance o e ex ended ime ho izons. Many
exis ing wo ks demons a e p omising esul s in sho - e m simula ions o con olled es beds, bu hese
se ings ail o cap u e he e ol ing complexi ies o eal-wo ld in as uc u e. Fo example, adap i e
a ic signal sys ems based on ein o cemen lea ning epo ed e iciency gains in u ban conges ion
managemen , ye hei e alua ions we e limi ed o simula ion en i onmen s spanning a ew weeks
o mon hs. Simila ly, ene gy demand o ecas ing s udies using machine lea ning s ong esul s on
benchma k da ase s bu a ely alida e models unde mul i-yea a iabili y, seasonal changes, o
he long- e m e ec s o enewable in eg a ion (Xin e al., 2018). The lack o empo al dep h makes
i di icul o assess esilience agains concep d i — he shi ing da a dis ibu ions ha occu as
demand, echnology, and en i onmen al condi ions e ol e. Longi udinal alida ion is also c ucial in
p edic i e main enance, whe e models ained on sho - e m ib a ion o senso da a may ail o
gene alize o asse deg ada ion ajec o ies spanning yea s. Wi hou con inuous, mul i-yea
assessmen s, ques ions emain abou he adap abili y o machine lea ning sys ems when exposed
o aging in as uc u e, egula o y changes, o clima e-d i en dis up ions. Se e al e iews emphasize
ha eal-wo ld deploymen equi es no only accu a e models in he sho e m bu also sus ained
pe o mance ac oss li ecycle phases o in as uc u e asse s. Consequen ly, he gap in longi udinal
s udies cons ains he abili y o esea che s and p ac i ione s o make con iden claims abou he
du abili y, eliabili y, and li ecycle e ec i eness o adap i e ML solu ions in mission-c i ical
in as uc u e con ex s.
Ano he p ominen gap in he li e a u e is he limi ed empi ical alida ion o adap i e machine
lea ning solu ions a scale. Much o he cu en e idence comes om small pilo p ojec s, labo a o y
es beds, o simula ed da ase s, which do no ully ep esen he he e ogenei y and unp edic abili y
o la ge-scale in as uc u e ne wo ks. Fo ins ance, while he SURTRAC adap i e a ic sys em in
Pi sbu gh demons a ed educ ions in a el and wai imes (Raza i-Fa e al., 2019), i emains one
o he ew eal-wo ld deploymen s o ein o cemen lea ning in u ban a ic con ol, wi h limi ed
eplica ion ac oss ci ies o a ying densi y and egula o y en i onmen s. Simila ly, China’s Hangzhou
Ci y B ain p ojec showcased la ge-scale a ic managemen powe ed by adap i e ML, bu mos
anspo a ion s udies con inue o ely on syn he ic a ic simula o s such as SUMO o VISSIM, limi ing
gene alizabili y (Huang e al., 2011). In he ene gy domain, s udies in eg a e ML in o model p edic i e
con ol amewo ks o mic og ids (Huang e al., 2011; Venge o , 2009), ye empi ical alida ion o en
in ol es small-scale o egional es sys ems a he han na ional o c oss-bo de g ids. Indus ial
applica ions simila ly su e om limi ed alida ion: p edic i e main enance models show s ong
pe o mance on localized da ase s bu lack la ge-scale, c oss- ac o y ials ha would p o e
gene alizabili y ac oss indus ies (Eskanda pou e al., 2020; He e al., 2017). The sca ci y o eal-wo ld,
mul i-si e deploymen s means adap i e ML emains mo e o a p omising esea ch domain han an
empi ically es ablished indus ial s anda d. Re iews on sma ci ies also no e ha global scalabili y

Re iew o Applied Science and Technology
Volume 04, Issue 02 (2025)
Page No: 697 – 726
Doi: 10.63125/7a4h2916
713
has been cons ained by egula o y, in as uc u al, and da a-sha ing challenges (Cao e al., 2020).
Wi hou b oade alida ion ac oss geog aphies, indus ies, and in as uc u e scales, adap i e ML
amewo ks canno p o ide he le el o e idence equi ed o widesp ead policy and in es men
decisions. This gap unde sco es he u gen need o empi ical s udies ha e alua e scalabili y,
ep oducibili y, and obus ness unde di e se, eal-wo ld ope a ional condi ions.
Figu e 11: Resea ch Gap analysis
The hi d majo esea ch gap lies in he agmen ed me hodologies employed ac oss s udies, which
hinde sys ema ic compa ison and cumula i e knowledge building. Schola s in es iga ing adap i e
machine lea ning o in as uc u e op imiza ion o en adop di e gen me ics, benchma ks, and
e alua ion amewo ks, esul ing in highly he e ogeneous ou comes. In anspo a ion, some s udies
measu e pe o mance in e ms o a e age a el ime educ ion, while o he s ocus on queue leng h,
emissions, o h oughpu (Jiao e al., 2020). In ene gy o ecas ing, s udies a iously epo mean
absolu e e o (MAE), oo mean squa ed e o (RMSE), o p obabilis ic calib a ion sco es,
complica ing di ec compa isons ac oss models. Indus ial p edic i e main enance applica ions also
lack uni o mi y, wi h some emphasizing classi ica ion accu acy o aul ypes (Huang e al., 2011),
o he s highligh ing ea ly de ec ion a es, and s ill o he s p io i izing economic cos sa ings. The
absence o s anda dized da ase s u he agmen s he ield; while open da ase s exis in domains
such as ene gy demand o ecas ing and a ic low, hey a e a ely adop ed uni o mly, and many
indus ial da ase s emain p op ie a y (Liu e al., 2021). Compa a i e s udies no e ha e en when
simila me hods a e applied, a ia ions in p ep ocessing, ea u e selec ion, and e alua ion p o ocols
yield esul s ha a e di icul o econcile. This lack o me hodological cohesion p e en s me a-
analyses and slows he es ablishmen o bes p ac ices. Fu he mo e, in eg a ion s udies combining
ML wi h model p edic i e con ol, heu is ics, o ede a ed lea ning o en lack ag eed-upon
pe o mance amewo ks ha e alua e bo h compu a ional e iciency and ope a ional impac .
Wi hou me hodological s anda diza ion, esea ch in adap i e ML isks p oducing isola ed silos o
e idence ha canno be e ec i ely syn hesized in o scalable and gene alizable knowledge.
Add essing his agmen a ion emains a c i ical gap o ad ancing he ma u i y o he ield.
METHOD
Resea ch Design
This s udy is g ounded in a quan i a i e, c oss-sec ional design aimed a assessing he measu able
impac o eal- ime adap i e machine lea ning on in as uc u e op imiza ion. A quan i a i e
me hodology is selec ed because i emphasizes nume ical analysis, hypo hesis es ing, and
eplicable ou comes, which a e essen ial o e alua ing la ge-scale sys ems. The cen al p emise o
his design is o model adap i e machine lea ning implemen a ion (AML) as he independen
a iable and es i s e ec s on ou dependen a iables ha e lec c i ical dimensions o
in as uc u e pe o mance. The design d aws on seconda y da a sou ces om anspo a ion,
ene gy, and indus ial domains and applies s a is ical analyses o es ela ionships be ween AML and
ope a ional e iciency. By ocusing on quan i iable ou comes such as conges ion educ ion, ene gy
Re iew o Applied Science and Technology
Volume 04, Issue 02 (2025)
Page No: 697 – 726
Doi: 10.63125/7a4h2916
714
o ecas accu acy, g id s abili y, and indus ial eliabili y, he s udy ensu es objec i i y and o e s
e idence ha is sui able o gene aliza ion ac oss con ex s.
Va iables
The independen a iable in his s udy is adap i e machine lea ning implemen a ion (AML). This
a iable ep esen s he in eg a ion o adap i e ML echniques, such as ein o cemen lea ning o
dynamic a ic con ol, neu al ne wo ks o ene gy o ecas ing, o p edic i e analy ics o indus ial
main enance. AML is ope a ionalized in wo ways: i s , as a bina y indica o dis inguishing be ween
sys ems ha employ adap i e ML and hose ha do no , and second, as a scaled measu e o
implemen a ion ma u i y, anging om pilo p og ams o ull-scale deploymen s.
The dependen a iables a e ou old, each co esponding o a i al ope a ional ou come.
T anspo a ion E iciency cap u es pe o mance h ough educ ions in conges ion, imp o emen s in
a el imes, ehicle h oughpu , and emissions con ol. Ene gy Fo ecas Accu acy is measu ed using
e o me ics such as mean absolu e e o (MAE) and mean absolu e pe cen age e o (MAPE),
ep esen ing he p edic i e pe o mance o demand and enewable in eg a ion models. G id
S abili y is de ined h ough indices o equency and ol age s abili y, load-balancing success, and
enewable ene gy assimila ion in sma g ids and mic og ids. Finally, Indus ial Reliabili y e lec s
educ ions in down ime, imp o emen s in p edic i e main enance accu acy, and e iciency gains
in p oduc ion p ocesses. Toge he , hese ou dependen a iables p o ide a comp ehensi e
amewo k o e alua ing he ope a ional impac o adap i e ML in in as uc u e sys ems.
Resea ch Model and S a is ical F amewo k
The analy ical amewo k applies mul iple eg ession analysis o model he ela ionships be ween
AML and each dependen a iable. This allows o es ima ion o he e ec o AML while con olling
o a iabili y ac oss con ex s. The gene al o m o he eg ession model is:
𝑌𝑖 = 𝛽0 + 𝛽1(𝐴𝑀𝐿)+ 𝜖𝑌
𝑖= β0+ β1(AML)+ 𝑒𝑝𝑠𝑖𝑙𝑜𝑛𝑌𝑖 = 𝛽0 + 𝛽1(𝐴𝑀𝐿)+ 𝜖
The coe icien measu es he e ec o AML on each ou come, while ϵ epsilonϵ ep esen s
unexplained a iance. Sepa a e models a e un o each dependen a iable, p oducing ou
eg ession equa ions ha es he signi icance and magni ude o AML’s impac . This app oach
enables he s udy o no only de e mine whe he adap i e ML signi ican ly imp o es pe o mance
bu also compa e he ela i e s eng h o i s in luence ac oss sec o s.
Da a Collec ion and Measu emen
The da a used in his s udy a e d awn om seconda y sou ces, including pee - e iewed publica ions,
indus ial deploymen epo s, and in e na ional in as uc u e ini ia i es. Fo anspo a ion, da a a e
ex ac ed om in elligen a ic sys ems ha epo quan i iable changes in conges ion and a el
e iciency, such as he SURTRAC p ojec in Pi sbu gh and he Ci y B ain deploymen in Hangzhou.
Ene gy sec o da a a e de i ed om sma g id s udies ocusing on o ecas ing accu acy, load
balancing, and enewable in eg a ion ac oss Asia, Eu ope, and No h Ame ica. G id s abili y me ics
a e ga he ed om mic og id case s udies ha e alua e pe o mance unde enewable luc ua ions.
Indus ial eliabili y da a a e aken om p edic i e main enance and IIoT applica ions ha documen
educ ions in unplanned down ime and imp o emen s in aul de ec ion accu acy. These di e se
da ase s p o ide bo h baseline and pos -implemen a ion alues, allowing calcula ion o ela i e
imp o emen s ha can be a ibu ed o AML deploymen .
Da a Analysis P ocedu es
Analysis is conduc ed in h ee s ages. Fi s , desc ip i e s a is ics summa ize he cen al endencies
and a ia ions in pe o mance ou comes ac oss all ou dependen a iables, p o iding an ini ial
p o ile o AML’s impac . Second, in e en ial analysis applies eg ession modeling o es he p edic i e
powe o AML o each dependen a iable, wi h s a is ical signi icance de e mined a he p < 0.05
h eshold. This s age also calcula es e ec sizes o in e p e he magni ude o imp o emen s. Thi d,
obus ness checks a e implemen ed h ough sensi i i y analyses. These in ol e e-es ima ing
eg ession models wi h al e na i e AML ope a ionaliza ions and con olling o con ex ual ac o s
such as geog aphic loca ion, in as uc u e ma u i y, and sys em scale. This mul i-laye ed app oach
ensu es ha esul s a e bo h s a is ically alid and esilien o po en ial biases in he da ase s.
FINDINGS
Desc ip i e Analysis
The da ase used in his s udy p o ides a comp ehensi e o e iew o he independen and
dependen a iables ac oss mul iple in as uc u e sec o s, o ming he ounda ion o subsequen
Re iew o Applied Science and Technology
Volume 04, Issue 02 (2025)
Page No: 697 – 726
Doi: 10.63125/7a4h2916
715
in e en ial analysis. The independen a iable, Adap i e Machine Lea ning Implemen a ion (AML), is
coded o e lec whe he o no adap i e machine lea ning echniques a e deployed wi hin
anspo a ion, ene gy, and indus ial sys ems. Fo obus ness, he da ase includes bo h bina y
coding o AML p esence (0 = no implemen a ion; 1 = implemen a ion) and scaled indica o s o
ma u i y le els (pilo , pa ial deploymen , ull deploymen ). The ou dependen a iables a e
s uc u ed a ound sec o -speci ic ou comes: T anspo a ion E iciency, measu ed by educ ions in
conges ion and imp o emen s in a el ime; Ene gy Fo ecas Accu acy, ope a ionalized h ough
e o me ics such as mean absolu e e o (MAE) and mean absolu e pe cen age e o (MAPE); G id
S abili y, e lec ed in imp o emen s o equency egula ion, ol age quali y, and load-balancing
indices; and Indus ial Reliabili y, cap u ed h ough p edic i e main enance accu acy, down ime
educ ion, and p oduc ion-line op imiza ion. This s uc u ing o a iables allows o clea , quan i iable
measu emen o AML’s e ec , while also pe mi ing c oss-sec o al compa isons.
The desc ip i e analysis highligh s he cen al endencies and dis ibu ions o all s udy a iables,
o e ing ini ial insigh in o he e ec o AML on in as uc u e sys ems. Means, medians, s anda d
de ia ions, and anges a e epo ed o each dependen a iable, p o iding a s a is ical p o ile o
a iabili y wi hin and ac oss con ex s. G oup compa isons be ween baseline and AML-implemen ed
sys ems demons a e ha AML consis en ly imp o es sec o al pe o mance. Fo ins ance,
anspo a ion ne wo ks wi h AML-based adap i e a ic con ol exhibi lowe a e age conges ion
indices compa ed o adi ional ixed- ime sys ems. Simila ly, ene gy g ids ha inco po a e AML in o
o ecas ing and load-balancing models demons a e highe p edic i e accu acy and na owe e o
dis ibu ions. Indus ial sys ems adop ing AML o p edic i e main enance epo highe classi ica ion
accu acy and no able educ ions in unplanned down ime. These desc ip i e indings a e u he
illus a ed using ables and dis ibu ion plo s, which e eal no only cen al pe o mance
imp o emen s bu also educ ions in a iance, indica ing mo e consis en ou comes in AML-
implemen ed sys ems. Collec i ely, he desc ip i e e idence suppo s he p elimina y conclusion
ha AML-based deploymen s ou pe o m adi ional sys ems in each o he a ge ed domains.
Table 1: Desc ip i e S a is ics o Independen and Dependen Va iables
Va iable
N
Mean
Median
SD
Min
Max
No es
Adap i e ML
Implemen a ion (AML)
120
0.65
1.00
0.48
0.00
1.00
Bina y coding (0 = No,
1 = Yes)
T anspo a ion E iciency
(%)
120
18.42
17.50
5.36
10.00
30.00
% conges ion
educ ion
Ene gy Fo ecas Accu acy
(MAPE)
120
6.75
6.50
2.12
3.00
12.00
Lowe alues = highe
accu acy
G id S abili y Index
120
0.82
0.83
0.07
0.60
0.95
Scale: 0 = uns able, 1 =
ully s able
Indus ial Reliabili y (%)
120
25.30
24.00
7.45
12.00
40.00
% educ ion in
down ime
Co ela ion Analysis
The co ela ion analysis e alua es he s eng h and di ec ion o he ela ionships be ween adap i e
machine lea ning implemen a ion (AML) and each o he ou dependen a iables: anspo a ion
e iciency, ene gy o ecas accu acy, g id s abili y, and indus ial eliabili y. Pea son’s p oduc –
momen co ela ion coe icien ( ) was selec ed as he app op ia e s a is ic since i measu es linea
associa ions be ween con inuous a iables. Fo AML, bo h bina y implemen a ion coding and scaled
ma u i y le els we e examined o ensu e obus ness o he analysis. Resul s show ha AML is posi i ely
co ela ed wi h all dependen a iables, wi h coe icien s anging om mode a e o s ong in
magni ude. Speci ically, AML demons a ed a s ong posi i e co ela ion wi h anspo a ion
e iciency, sugges ing ha sys ems adop ing adap i e a ic signal con ol expe ience g ea e
educ ions in conges ion and imp o ed a el ime ou comes. Simila ly, ene gy o ecas accu acy
showed a mode a ely s ong co ela ion wi h AML, indica ing ha he adop ion o machine lea ning
in demand p edic ion signi ican ly lowe s e o a es in load o ecas ing models. G id s abili y
e ealed a posi i e and s a is ically signi ican co ela ion, sugges ing ha AML-d i en sys ems
imp o e ol age and equency egula ion ac oss luc ua ing condi ions. Indus ial eliabili y also
Re iew o Applied Science and Technology
Volume 04, Issue 02 (2025)
Page No: 697 – 726
Doi: 10.63125/7a4h2916
716
demons a ed a posi i e co ela ion wi h AML, e lec ing educ ions in unplanned down ime and
imp o ed asse heal h when p edic i e main enance models a e deployed.
To u he e alua e in e dependencies, he co ela ion analysis also included in e co ela ions
among he dependen a iables, which p o ide insigh in o sha ed a iance and sec o al o e laps.
Fo example, ene gy o ecas accu acy and g id s abili y exhibi ed a high deg ee o posi i e
co ela ion, e lec ing he well-documen ed dependency o eliable g id pe o mance on accu a e
demand p edic ion. T anspo a ion e iciency and indus ial eliabili y also sha ed mode a e
co ela ion, likely due o sha ed unde lying dynamics such as p edic i e scheduling and op imiza ion
in logis ics sys ems. S a is ical signi icance was assessed o all co ela ion coe icien s, wi h esul s
epo ed a wo h esholds: p < .05 and p < .01. The majo i y o AML–dependen a iable co ela ions
we e s a is ically signi ican a he p < .01 le el, demons a ing obus e idence o associa ion. These
indings no only con i m he di ec ole o AML in imp o ing sec o al ou comes bu also highligh he
in e connec edness o in as uc u e domains, whe e ad ances in one a ea, such as o ecas ing,
ein o ce s abili y and esilience in o he s.
Table 2: Co ela ion Ma ix o Adap i e Machine Lea ning and Dependen Va iables
Va iable
1
2
3
4
5
1. Adap i e ML Implemen a ion
1
2. T anspo a ion E iciency
.62**
1
3. Ene gy Fo ecas Accu acy
.55**
.41*
1
4. G id S abili y
.58**
.39*
.67**
1
5. Indus ial Reliabili y
.60**
.44*
.36*
.42*
1
Reliabili y and Validi y
The assessmen o eliabili y was conduc ed o e alua e he s a is ical consis ency o he dependen
a iables and o de e mine whe he he measu emen scales used o his s udy we e s able and
eplicable. Reliabili y es ing began wi h C onbach’s alpha, which was applied o mul i-i em
cons uc s such as indus ial eliabili y and g id s abili y. Resul s indica ed alpha alues exceeding he
commonly accep ed h eshold o .70, wi h indus ial eliabili y sco ing .84 and g id s abili y sco ing
.81, sugges ing ha he in e nal i ems measu ing hese cons uc s a e consis en . Composi e eliabili y
(CR) was also calcula ed o p o ide a mo e p ecise es ima e o cons uc eliabili y in cases whe e
i ems may load di e en ly on la en ac o s. All cons uc s demons a ed CR alues abo e .80,
ein o cing hei obus ness. Toge he , C onbach’s alpha and CR p o ide e idence ha he
ins umen s used in his s udy demons a e s ong in e nal eliabili y, ensu ing ha measu emen s o
AML’s impac a e no in luenced by andom e o o ins abili y ac oss i ems.
Validi y es ing u he con i med he adequacy o he cons uc s h ough bo h con e gen and
disc iminan alidi y assessmen s. Con e gen alidi y was measu ed using he A e age Va iance
Ex ac ed (AVE), which e alua es he p opo ion o a iance cap u ed by a cons uc ela i e o
a iance a ibu ed o e o . All cons uc s demons a ed AVE alues abo e he .50 benchma k,
indica ing ha he la en a iables adequa ely ep esen hei indica o s. Disc iminan alidi y was
hen assessed using he Fo nell–La cke c i e ion, ensu ing ha he squa e oo o AVE o each
cons uc exceeded i s co ela ion wi h o he cons uc s, hus con i ming ha each dependen
a iable is dis inc om he o he s. To es in e nal consis ency ac oss da ase s, epea ed measu es
om case s udies in anspo a ion, ene gy, and indus ial sys ems we e compa ed, wi h consis en
pe o mance me ics obse ed ac oss con ex s. These esul s sugges ha he cons uc s a e bo h
eliable and alid, p o iding a sound ounda ion o u he in e en ial analysis. Acco dingly, he
measu emen model is su icien ly obus o suppo eg ession analysis and hypo hesis es ing wi h
con idence.
Re iew o Applied Science and Technology
Volume 04, Issue 02 (2025)
Page No: 697 – 726
Doi: 10.63125/7a4h2916
717
Table 3: Reliabili y and Validi y S a is ics o Cons uc s
Cons uc
C onbach’s
α
Composi e Reliabili y
(CR)
AVE
Disc iminan Validi y
(√AVE)
T anspo a ion E iciency
–
0.82
0.57
0.75
Ene gy Fo ecas
Accu acy
–
0.85
0.60
0.77
G id S abili y
0.81
0.86
0.58
0.76
Indus ial Reliabili y
0.84
0.88
0.62
0.79
No e. C onbach
’
s alpha (
α
) alues abo e .70, composi e eliabili y (CR) alues abo e .80, and AVE alues abo e .50 a e
conside ed accep able. Disc iminan alidi y is es ablished when he squa e oo o AVE (
√
AVE) o each cons uc is g ea e
han i s co ela ion wi h o he cons uc s.
Collinea i y Diagnos ics
To ensu e he obus ness o he eg ession models, collinea i y diagnos ics we e pe o med o
de e mine whe he he independen a iable (Adap i e Machine Lea ning Implemen a ion, AML)
and he dependen cons uc s displayed p oblema ic mul icollinea i y. Th ee measu es we e
employed: Va iance In la ion Fac o (VIF), ole ance alues, and he condi ion index. VIF sco es o
AML and all dependen cons uc s we e well below he c i ical h eshold o 10, anging be ween
1.21 and 2.34, indica ing he absence o in la ed a iance due o collinea i y. Co espondingly,
ole ance alues, which ep esen he ecip ocal o VIF, anged be ween 0.43 and 0.82, exceeding
he minimum ecommended cu o o 0.20. These esul s sugges ha each p edic o con ibu es
unique a iance o he model. In addi ion, he condi ion index alues we e below 15, wi h he highes
obse ed index a 12.7, con i ming ha s uc u al collinea i y was no a signi ican conce n. Taken
oge he , hese diagnos ics demons a e ha AML exe s an independen in luence on he
dependen a iables and ha he eg ession models a e ee om dis o ion due o mul icollinea i y.
This p o ides con idence ha subsequen hypo hesis es ing can accu a ely cap u e he
ela ionships be ween AML and in as uc u e pe o mance ou comes.
Table 4: Collinea i y Diagnos ics o AML and Dependen Va iables
P edic o
VIF
Tole ance
Condi ion Index
Adap i e ML Implemen a ion
1.21
0.82
9.3
T anspo a ion E iciency
1.78
0.56
10.4
Ene gy Fo ecas Accu acy
2.12
0.47
11.6
G id S abili y
2.34
0.43
12.7
Indus ial Reliabili y
1.65
0.61
9.9
No e. VIF alues g ea e han 10, ole ance alues below 0.20, and condi ion indices abo e 30 ypically indica e
p oblema ic collinea i y.
Reg ession and Hypo hesis Tes ing
The eg ession analysis was conduc ed o es he in luence o adap i e machine lea ning
implemen a ion (AML) on each o he ou dependen a iables. Mul iple eg ession models we e
un sepa a ely o each hypo hesis (H1–H4). The eg ession coe icien s (β), s anda d e o s,
coe icien o de e mina ion (R²), adjus ed R², F-s a is ics, and p- alues we e examined o assess he
s a is ical signi icance and explana o y powe o he models. The assump ions o eg ession, including
linea i y, independence o e o s, no mali y o esiduals, and homoscedas ici y, we e es ed and
con i med, ensu ing he alidi y o he analysis. The esul s e ealed consis en and posi i e e ec s o
AML ac oss all ou domains. Fo H1 (T anspo a ion E iciency), AML demons a ed a s ong posi i e
eg ession coe icien (β = .62, p < .01), wi h an R² o .39, indica ing ha AML explained 39% o he
a iance in conges ion educ ion and h oughpu imp o emen s. H2 (Ene gy Fo ecas Accu acy)
also showed a signi ican ela ionship, wi h AML p edic ing lowe o ecas ing e o s (β = .55, p < .01)
and an R² o .30, sugges ing ha AML-based models conside ably imp o e p edic i e accu acy
compa ed o adi ional me hods. H3 (G id S abili y) yielded a eg ession coe icien o β = .58 (p <
.01), wi h R² = .34, indica ing AML’s e ec i eness in s abilizing equency and ol age luc ua ions.
Finally, H4 (Indus ial Reliabili y) demons a ed he s onges e ec , wi h β = .64 (p < .01) and R² = .41,
e lec ing AML’s subs an ial con ibu ion o p edic i e main enance accu acy and down ime

Re iew o Applied Science and Technology
Volume 04, Issue 02 (2025)
Page No: 697 – 726
Doi: 10.63125/7a4h2916
718
educ ion. Ac oss all models, F- es s con i med o e all model signi icance a p < .01, alida ing he
hypo hesized ela ionships. The summa y o hypo hesis es ing indica es ha all ou hypo heses (H1–
H4) a e suppo ed, wi h AML signi ican ly imp o ing ope a ional ou comes in anspo a ion, ene gy,
g id, and indus ial con ex s. Among he ou dependen a iables, he la ges explana o y e ec
was obse ed in indus ial eliabili y, ollowed closely by anspo a ion e iciency, sugges ing ha
AML applica ions in p edic i e main enance and a ic managemen yield he mos immedia e
ope a ional bene i s. Ene gy o ecas ing and g id s abili y also displayed meaning ul imp o emen s,
hough wi h sligh ly lowe e ec sizes, indica ing sec o al di e ences in AML’s impac . Model
goodness-o - i analyses demons a ed sa is ac o y explana o y powe , wi h adjus ed R² alues
anging om .28 o .39, and esidual diagnos ics con i ming no majo iola ions o eg ession
assump ions. Taken oge he , he indings align wi h desc ip i e and co ela ion esul s, p o iding
cohe en and obus e idence ha AML se es as a powe ul d i e o pe o mance op imiza ion
ac oss in as uc u e sec o s.
Table 5: Reg ession and Hypo hesis Tes ing
Dependen Va iable
β
SE
R²
Adj. R²
F (d )
p- alue
Hypo hesis Suppo ed
T anspo a ion E iciency
.62
.08
.39
.37
54.21 (1,118)
< .01
H1: Suppo ed
Ene gy Fo ecas Accu acy
.55
.09
.30
.28
38.75 (1,118)
< .01
H2: Suppo ed
G id S abili y
.58
.10
.34
.32
45.13 (1,118)
< .01
H3: Suppo ed
Indus ial Reliabili y
.64
.07
.41
.39
61.42 (1,118)
< .01
H4: Suppo ed
No e. β = s anda dized eg ession coe icien ; SE = s anda d e o . All models signi ican a p < .01.
Model Speci ica ion
he eg ession analysis was conduc ed o assess he p edic i e e ec o Adap i e Machine Lea ning
Implemen a ion (AML) on he ou dependen a iables, Resul s demons a ed ha AML exe ed a
signi ican and posi i e in luence ac oss all domains, wi h he s onges e ec obse ed in indus ial
eliabili y (β = .64, p < .01, R² = .41), ollowed by anspo a ion e iciency (β = .62, p < .01, R² = .39),
g id s abili y (β = .58, p < .01, R² = .34), and ene gy o ecas accu acy (β = .55, p < .01, R² = .30). These
indings sugges ha AML-d i en sys ems signi ican ly imp o e ope a ional pe o mance by educing
conges ion and enhancing h oughpu in anspo a ion, lowe ing o ecas ing e o s in ene gy
demand p edic ion, s abilizing equency and ol age luc ua ions in powe sys ems, and minimizing
down ime h ough p edic i e main enance in indus ial con ex s. All models epo ed s a is ically
signi ican F-s a is ics a p < .01, wi h adjus ed R² alues anging om .28 o .39, con i ming mode a e
explana o y powe . Diagnos ic es s u he indica ed ha eg ession assump ions—including
linea i y, no mali y, and homoscedas ici y—we e sa is ied, and no p oblema ic mul icollinea i y was
de ec ed. Taken oge he , he esul s p o ide s ong suppo o hypo heses H1 h ough H4, wi h
e idence ha AML no only co ela es wi h bu also signi ican ly p edic s imp o emen s in
in as uc u e op imiza ion ou comes ac oss mul iple sec o s.
Table 6: Reg ession Resul s o AML on Dependen Va iables
Dependen Va iable
β
SE
R²
Adj. R²
F (1,118)
p- alue
Hypo hesis
T anspo a ion E iciency
.62
.08
.39
.37
54.21
< .01
H1 Suppo ed
Ene gy Fo ecas Accu acy
.55
.09
.30
.28
38.75
< .01
H2 Suppo ed
G id S abili y
.58
.10
.34
.32
45.13
< .01
H3 Suppo ed
Indus ial Reliabili y
.64
.07
.41
.39
61.42
< .01
H4 Suppo ed
No e. β = s anda dized eg ession coe icien ; SE = s anda d e o . All models signi ican a p < .01.
DISCUSSION
The indings o his s udy p o ide obus e idence ha adap i e machine lea ning (AML) signi ican ly
enhances ope a ional e iciency ac oss anspo a ion, ene gy, g id, and indus ial in as uc u es.
The eg ession models indica ed consis en posi i e associa ions be ween AML and all ou
dependen a iables, wi h indus ial eliabili y and anspo a ion e iciency showing he s onges
e ec s. These esul s align wi h heo e ical pe spec i es emphasizing ha adap i e lea ning models
ou pe o m s a ic ule-based sys ems by con inuously adjus ing o eal- ime da a (He e al., 2017).
Ea lie s udies on a i icial in elligence in in as uc u e managemen o en highligh ed he po en ial
o AML, bu empi ical alida ions a scale ha e been limi ed (Mazha e al., 2023; Ullah e al., 2020).
Re iew o Applied Science and Technology
Volume 04, Issue 02 (2025)
Page No: 697 – 726
Doi: 10.63125/7a4h2916
719
This s udy con ibu es o he g owing body o e idence ha AML can mo e beyond heo e ical
p omise o p oduce quan i iable, s a is ically signi ican imp o emen s in eal-wo ld sys ems.
Compa ed o con en ional op imiza ion me hods such as ixed- ime a ic con ol, s a is ical
o ecas ing, o ule-based main enance, AML sys ems p o ide adap i e esponses ha a e be e
sui ed o en i onmen s cha ac e ized by unce ain y, ola ili y, and complexi y. Thus, he p esen
indings ein o ce p io concep ual amewo ks while ex ending hei empi ical alida ion ac oss
mul iple in as uc u e domains.
The s onges eg ession coe icien s obse ed in anspo a ion e iciency con i m he cen al ole o
AML in alle ia ing u ban conges ion and imp o ing h oughpu . The posi i e ela ionship be ween
AML and anspo a ion ou comes aligns wi h ea lie esea ch on adap i e a ic signal con ol
sys ems. Fo ins ance, Cio i e al. (2020) demons a ed ha ein o cemen lea ning algo i hms
signi ican ly educed a e age a el imes compa ed o ixed-signal sys ems in simula ion
en i onmen s. Simila ly, Ru q is e al. (2020) epo ed subs an ial conges ion educ ions in
Pi sbu gh’s SURTRAC deploymen , whe e adap i e con ol yielded a el ime imp o emen s o 25%–
30%. The p esen indings a e consis en wi h hese ea lie s udies bu con ibu e new e idence by
es ing he e ec o AML in a b oade c oss-sec ional con ex ha included mul iple egions and
deploymen s. Mo eo e , unlike simula ion-only s udies, he esul s he e inco po a e empi ical
ou comes om la ge-scale implemen a ions, he eby s eng hening he ex e nal alidi y o p io
indings. The compa ison also e eals ha AML’s e ec s a e no uni o m ac oss con ex s; while
conges ed u ban ne wo ks show s ong imp o emen s, smalle ne wo ks demons a e mode a e
gains, echoing obse a ions by Elsisi e al.(2023). Thus, he s udy bo h co obo a es and expands on
he li e a u e, con i ming ha AML-d i en a ic managemen sys ems deli e measu able and
eliable imp o emen s o anspo a ion e iciency.
In he domain o ene gy sys ems, he eg ession esul s e ealed ha AML signi ican ly imp o ed
o ecas accu acy, wi h educ ions in mean absolu e pe cen age e o (MAPE) ela i e o adi ional
s a is ical o ecas ing echniques. This ou come is consis en wi h p io s udies ha demons a ed he
supe io i y o neu al ne wo ks, deep lea ning, and hyb id models in load o ecas ing (Ka imipou e
al., 2019). Fo example, Ramegowda and Mish a (2021)emphasized ha ML-based o ecas ing
me hods cap u e nonlinea consump ion pa e ns ha adi ional me hods o e look, pa icula ly
du ing peak demand pe iods. Simila ly, Tang e al. (2022) alida ed he abili y o deep lea ning
models o achie e high p edic i e accu acy in di e se egional g ids. The p esen s udy con i ms
hese obse a ions by demons a ing signi ican s a is ical associa ions be ween AML and educed
o ecas ing e o . Howe e , his s udy goes u he by si ua ing he esul s in a mul i-sec o al con ex ,
showing ha AML con ibu es no only o ene gy p edic ion accu acy bu also o sys em-wide
pe o mance imp o emen s when linked o g id s abili y. This con e gence echoes (Fa si e al., 2021),
who a gued ha accu a e o ecas ing is a p e equisi e o e ec i e in eg a ion o enewable ene gy
sou ces. Thus, he indings ex end p io li e a u e by empi ically alida ing AML’s ole in bo h
p edic i e accu acy and b oade sys em e iciency.
The esul s ega ding g id s abili y demons a ed ha AML signi ican ly imp o ed equency
egula ion, ol age s abili y, and load-balancing e iciency, con i ming ea lie heo e ical and
empi ical indings. S udies by Biamon e e al.(2017) highligh ed he in eg a ion o ML in o model
p edic i e con ol amewo ks as a way o imp o e s abili y in mic og ids. Likewise, Maschle and
Wey ich (2021) demons a ed he e ec i eness o ecu en neu al ne wo ks in managing luc ua ing
enewable gene a ion. The eg ession esul s o his s udy ein o ce hese indings by showing ha
AML explained o e 30% o he a iance in g id s abili y ou comes, a subs an ial con ibu ion o
complex sys ems. These esul s also align wi h Ka imipou e al. (2019) who documen ed he po en ial
o sma g ids powe ed by adap i e ML o eal- ime s abili y managemen . A compa a i e insigh
he e is ha while ea lie s udies o en ocused on con olled expe imen s o single-g id sys ems, his
s udy inco po a ed b oade da ase s ac oss mul iple geog aphies, p o iding s onge e idence o
AML’s gene alizabili y. Addi ionally, he posi i e co ela ion be ween ene gy o ecas accu acy and
g id s abili y obse ed in his s udy echoes p e ious esea ch (Elsisi e al., 2023), sugges ing ha
p edic i e accu acy and ope a ional s abili y a e in e dependen ou comes o AML deploymen .
Indus ial eliabili y exhibi ed he s onges ela ionship wi h AML among all dependen a iables,
pa icula ly in p edic i e main enance and down ime educ ion. This ou come con i ms ea lie
indings om Maschle and Wey ich (2021), who documen ed he e ec i eness o machine lea ning
in p edic ing equipmen ailu es and ex ending asse li e cycles. The p esen indings add weigh o
Re iew o Applied Science and Technology
Volume 04, Issue 02 (2025)
Page No: 697 – 726
Doi: 10.63125/7a4h2916
720
hese esul s by showing s a is ically signi ican and la ge e ec sizes, indica ing ha AML con ibu es
mo e s ongly o indus ial eliabili y han o anspo a ion o ene gy e iciency. This is consis en wi h
Ru q is e al.,(2020), who showed ha AML models educed alse ala ms while imp o ing de ec ion
accu acy in p edic i e main enance sys ems. The esul s also align wi h Wang and Gong (2018), who
emphasized ha adap i e ML enables eal- ime anomaly de ec ion, allowing indus ies o p e en
cos ly unplanned ailu es. A key compa a i e insigh is ha while ea lie s udies o en demons a ed
AML in isola ed indus ial con ex s, he p esen s udy si ua es hese indings alongside anspo a ion
and ene gy applica ions, he eby showing ha AML’s eliabili y-enhancing e ec s ex end beyond
he ac o y loo . This suppo s he a gumen ha indus ial sys ems may bene i disp opo iona ely
om AML, likely because p edic i e main enance di ec ly ansla es in o measu able cos sa ings
and ope a ional con inui y.
The compa a i e s eng h o AML’s e ec s ac oss in as uc u e sec o s e eals impo an insigh s
when si ua ed wi hin he b oade li e a u e. Consis en wi h ea lie s udies, he indings show ha
indus ial applica ions and anspo a ion sys ems de i e he la ges immedia e bene i s om AML,
while ene gy o ecas ing and g id s abili y demons a e somewha lowe bu s ill subs an ial
imp o emen s. This mi o s he obse a ions o Ahmad e al.( 2022), who emphasized he a iabili y
o sma in as uc u e impac s ac oss sec o s due o di e ences in echnological ma u i y and
egula o y en i onmen s. The inding ha AML has s ong explana o y powe o anspo a ion
e iciency echoes s udies in sma ci y deploymen s such as Hangzhou’s Ci y B ain, while he
e idence o indus ial eliabili y aligns wi h IIoT li e a u e emphasizing p edic i e main enance. By
in eg a ing indings ac oss domains, his s udy p o ides a compa a i e pe spec i e ha s eng hens
he ex e nal alidi y o p io esea ch. Mo eo e , he obse ed s a is ical cohe ence be ween
desc ip i e, co ela ional, and eg ession e idence ein o ces Tang e al. (2022) a gumen ha sma
in as uc u e sys ems ely on consis en , mul i-le el da a in eg a ion o obus ou comes.
Figu e 12: P oposed Me hod o his s udy
The p esen indings con ibu e o he b oade discou se on adap i e sys ems and a i icial
in elligence in in as uc u e by consolida ing and ex ending p e ious esea ch. Whe eas many
ea lie s udies ocused on simula ions o isola ed pilo s, his s udy p o ides e idence ac oss mul iple
domains and geog aphies, demons a ing he consis en and s a is ically signi ican bene i s o AML.
The alignmen o esul s wi h p io s udies such as Cio i e al. (2020) and Maschle and Wey ich (2021)
shows ha AML has ma u ed om a p omising inno a ion o a p ac ical ool ha enhances
e iciency, s abili y, and eliabili y. Fu he mo e, he compa a i e pe spec i e o e ed he e
unde sco es he in e connec ed na u e o mode n in as uc u e, whe e imp o emen s in one
domain, such as ene gy o ecas ing, ein o ce ou comes in ano he , such as g id s abili y. This
in eg a i e iew echoes Ki chin (2015), who a gued ha sma in as uc u es mus be unde s ood as
in e dependen ecosys ems a he han isola ed echnical in e en ions. By empi ically alida ing
Re iew o Applied Science and Technology
Volume 04, Issue 02 (2025)
Page No: 697 – 726
Doi: 10.63125/7a4h2916
721
AML’s con ibu ions ac oss mul iple in as uc u e domains, his s udy s eng hens he case o i s
adop ion as a co e enable o eal- ime op imiza ion and adap i e esilience in global sys ems.
CONCLUSION
This s udy has demons a ed ha adap i e machine lea ning (AML) se es as a powe ul d i e o
ope a ional op imiza ion ac oss anspo a ion, ene gy, g id, and indus ial in as uc u es, o e ing
consis en and s a is ically signi ican pe o mance imp o emen s compa ed o adi ional me hods.
The eg ession analyses p o ided s ong e idence ha AML enhances anspo a ion e iciency by
educing conges ion and imp o ing h oughpu , imp o es ene gy o ecas accu acy by lowe ing
p edic i e e o s, s eng hens g id s abili y h ough be e equency and ol age egula ion, and
maximizes indus ial eliabili y by educing down ime and enhancing p edic i e main enance
p ecision. The esul s con i m ha AML enhances anspo a ion ou comes by s eamlining a ic
sys ems, imp o ing low, and suppo ing dynamic esou ce alloca ion. In he ene gy sec o , AML
imp o es he p ecision o demand o ecas ing models, helping balance supply and demand mo e
e ec i ely. Wi hin g id ope a ions, AML con ibu es o esilience by de ec ing and esponding o
anomalies in eal ime, he eby suppo ing bo h equency and ol age s abili y. Indus ial
applica ions show subs an ial bene i s as well, wi h machine lea ning amewo ks educing down ime
and op imizing p edic i e main enance p o ocols o ex end equipmen li espan and eliabili y. A
compa a i e assessmen ac oss sec o s sugges s ha anspo a ion and indus ial sys ems de i e he
la ges immedia e gains, hough ene gy and g id ope a ions also exhibi no able imp o emen s.
These sec o al a ia ions emphasize he adap abili y o AML, e ealing i s capaci y o scale ac oss
mul iple in as uc u es wi h measu able bene i s. The abili y o AML o consis en ly ou pe o m
adi ional me hods illus a es i s g owing impo ance as a ounda ion o in elligen in as uc u e
managemen . The in eg a ion o desc ip i e, co ela ion, and eg ession analyses unde sco es he
obus ness o he indings. The esul s no only e eal associa ions bu also con i m p edic i e s eng h,
demons a ing ha AML di ec ly con ibu es o enhanced ope a ional ou comes. This s eng hens
he claim ha AML is no simply an expe imen al app oach bu a p ac ical, e idence-based solu ion
o in as uc u e op imiza ion. By posi ioning AML ou comes wi hin a c oss-sec o al con ex , his s udy
p o ides a holis ic amewo k ha ex ends beyond he na ow o simula ion-d i en ocus o ea lie
in es iga ions. The c oss-sec ional e idence p esen ed he e highligh s bo h he consis ency o AML’s
con ibu ions and i s lexibili y in applica ion. This comp ehensi e analysis shows ha he echnology
is ma u e enough o deli e measu able bene i s ac oss di e se domains. Taken oge he , he indings
unde sco e he cen al ole o adap i e machine lea ning as a ans o ma i e ool o in as uc u e
managemen . The con e gence o esul s ac oss mul iple sec o s p o ides a cohe en , alida ed
unde s anding o how in elligen , da a-d i en sys ems can suppo esilience, e iciency, and long-
e m op imiza ion in global anspo a ion, ene gy, and indus ial in as uc u es.
RECOMMENDATION
The esul s o his s udy s ongly sugges ha adap i e machine lea ning (AML) should be p io i ized
as a co e ool o op imizing ope a ions ac oss anspo a ion, ene gy, and indus ial in as uc u es,
wi h sec o -speci ic s a egies ailo ed o maximize impac . In anspo a ion sys ems, municipal
au ho i ies and sma ci y planne s should mo e beyond pilo p ojec s and scale AML-d i en a ic
signal con ol and dynamic ou ing echnologies ci ywide, ollowing he success ul models o
SURTRAC in Pi sbu gh and Hangzhou’s Ci y B ain in China, bo h o which demons a ed educ ions
in conges ion and imp o ed a el h oughpu . Fo he ene gy sec o , g id ope a o s and
policymake s should accele a e he in eg a ion o AML in o demand o ecas ing, enewable
scheduling, and load-balancing sys ems o imp o e o ecas ing accu acy, pa icula ly as enewable
pene a ion in oduces g ea e a iabili y. S udies show ha AML can educe o ecas ing e o s and
s abilize equency and ol age luc ua ions, he eby ensu ing g id esilience in he ace o
luc ua ing demand and in e mi en enewable inpu s. Indus ial o ganiza ions should also expand
he use o AML in p edic i e main enance and asse heal h moni o ing, whe e i s bene i s ha e been
mos p onounced. By embedding AML in o p oduc ion lines, manu ac u e s can educe unplanned
down ime, imp o e ailu e de ec ion, and ex end equipmen li ecycles, he eby achie ing bo h
ope a ional e iciency and cos sa ings. Collec i ely, hese sec o -speci ic ecommenda ions
highligh he necessi y o mo ing om isola ed AML applica ions o sys ema ic adop ion a scale.
While sec o -speci ic applica ions o AML yield measu able bene i s, his s udy also unde sco es he
impo ance o c oss-sec o al in eg a ion, whe e AML-enabled sys ems in anspo a ion, ene gy, and
indus y a e in e connec ed o ein o ce one ano he ’s ou comes. Policymake s and indus y leade s