A Systematic Literature Review of Advanced Machine Learning Techniques in Wireless Body Area Networks: Application, Challenges, and Future Directions

Recei ed 27 Oc obe 2025, accep ed 6 No embe 2025, da e o publica ion 10 No embe 2025,
da e o cu en e sion 19 No embe 2025.
Digi al Objec Iden i ie 10.1109/ACCESS.2025.3631230
A Sys ema ic Li e a u e Re iew o Ad anced
Machine Lea ning Techniques in Wi eless Body
A ea Ne wo ks: Applica ion, Challenges,
and Fu u e Di ec ions
ABDU IBRAHIM ADAMU 1, PRAVEEN KUMAR DONTA 2, (Senio Membe , IEEE),
DARMAWATY MOHD ALI 1, SOHAIL SARANG 3, (Senio Membe , IEEE),
GORAN M. STOJANOVIĆ 3, (Membe , IEEE), AND SUZI SEROJA SARNIN1
1Wi eless Communica ion Technology G oup (WiCOT), Facul y o Elec ical Enginee ing, Uni e si i Teknologi MARA (UiTM), Shah Alam, Selango 40450,
Malaysia
2Depa men o Compu e and Sys ems Sciences, Facul y o Social Sciences, S ockholm Uni e si y, 114 19 S ockholm, Sweden
3Facul y o Technical Sciences, Uni e si y o No i Sad, 21000 No i Sad, Se bia
Co esponding au ho : Da mawa y Mohd Ali ([email p o ec ed])
This wo k was suppo ed in pa by Eu opean Union’s Ho izon Eu ope Eu opean Inno a ion Council (EIC) 2023 Pa h inde Challenge
P og am unde G an 101161032, in pa by he Minis y o Highe Educa ion Malaysia (MOHE) o he Fundamen al Resea ch G an
Scheme (FRGS) unde G an FRGS/1/2023/TK07/UITM/02/29, and in pa by he Uni e si i Teknologi MARA (UiTM).
ABSTRACT The de elopmen o machine lea ning (ML) in wi eless body a ea ne wo ks (WBANs) has
made g ea p og ess in heal hca e moni o ing. The senso s a e made wea able o moni o he physiological
da a wi hou any ime lapse, con inuously. Impo an ly, issues such as ene gy consump ion, eliabili y o
senso da a, pa ien p i acy, and he need o anspa en , in e p e able models emain signi ican ba ie s o
he applica ion o his echnology in clinical en i onmen s. This pape p esen s a sys ema ic li e a u e e iew
(SLR) o pee - e iewed s udies published be ween 2017 and 2025, conduc ed in acco dance wi h P e e ed
Repo ing I ems o Sys ema ic Re iew and Me a-analysis (PRISMA) guidelines o ensu e igo and
ep oducibili y. F om a o al o 2,407 publica ions sc eened, 55 s udies me he inclusion c i e ia. This SLR
in es iga es how WBANs and ML ha e been applied o suppo anomaly de ec ion, ac i i y ecogni ion,
and ligh weigh da a ansmission o acili a e pe sonalized heal hca e. We compa e he s a ed pe o mance
me ics o ou di e en ML app oaches, supe ised lea ning, unsupe ised lea ning, ein o cemen lea ning
(RL), and hyb id app oaches, and map each o WBAN applica ion con ex s. To ackle hese challenges,
ad anced ML echniques a e s udied, including gene a i e a i icial in elligence (GAI), ede a ed lea ning
(FL), ligh weigh models, ep esen a ion lea ning, deep lea ning (DL), RL, and au oencode s. Mo eo e ,
low-la ency deep neu al ne wo ks (DNNs), edge compu ing, and eXplainable AI (XAI) echniques a e
ecommended o inc ease in e p e abili y and enable eal- ime decision-making. This syn hesis highligh s
pe sis en gaps in ene gy e iciency, scalabili y, p i acy p ese a ion, and s anda dized e alua ion. I also lays
ou a speci ic esea ch agenda o di ec u u e in es iga ions. We conclude ha ou c i ical issues p o ec ing
pa ien da a, ex ending ba e y li e, p ocessing da a quickly, and ensu ing accu a e and eliable senso
eadings mus be add essed i ML-powe ed wea ables a e o genuinely ans o m he heal hca e indus y.
INDEX TERMS Anomaly de ec ion, ene gy e iciency, machine lea ning, pe sonalized heal hca e, eal- ime
moni o ing, wi eless body a ea ne wo ks.
The associa e edi o coo dina ing he e iew o his manusc ip and
app o ing i o publica ion was Ayman El-Baz .
I. INTRODUCTION
Th ough wi eless body a ea ne wo ks (WBANs), wea -
able senso s a e ne wo ked o moni o and ansmi
VOLUME 13, 2025
2025 The Au ho s. This wo k is licensed unde a C ea i e Commons A ibu ion 4.0 License.
Fo mo e in o ma ion, see h ps://c ea i ecommons.o g/licenses/by/4.0/ 194729
A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
physiological da a om he human body, making WBANs a
apidly e ol ing heal hca e echnology. These ne wo ks a e
p oximi y-based and u ilize low- ange wi eless communica-
ion p o ocols, such as Blue oo h and Zigbee, o eal- ime
heal h moni o ing [1]. WBANs a e p ima ily applied in a -
ious a eas, including elemedicine, ch onic disease manage-
men , and exe cise moni o ing, assis ing heal hca e p o ide s
in wi elessly aking pa ien s’ i al signs and o he clinical
indica o s [2]. Wi h ecen ad ancemen s in senso echnol-
ogy, minia u iza ion, and ene gy ha es ing, WBAN de ices’
pe o mance, eliabili y, and accessibili y ha e signi ican ly
imp o ed, esul ing in hei g ea e usage in e e yday li e [3].
The inc easing p e alence o ch onic heal h condi ions has
inc eased he need o ongoing heal hca e moni o ing, which
has sped up he adop ion o WBAN in sec o s like spo s,
ehabili a ion, and elde ca e [4].
Recen ad ances in a i icial in elligence (AI) and machine
lea ning (ML) ha e caused unp eceden ed g ow h in mul i-
ple domains due o imp o ed algo i hms, highe p ocessing
powe , and he a ailabili y o gian da ase s. The wo ldwide
AI ma ke is p ojec ed o each $1.4 illion by 2029, a a
compound annual g ow h a e abo e 20% [5]. Ad ances in
deep lea ning (DL) algo i hms, such as ans o me models
in na u al language p ocessing and gene a i e ad e sa ial
ne wo ks (GANs) in image syn hesis, ha e u he ele a ed
AI’s po en ial. These echnologies ha e he po en ial o com-
ple ely ans o m heal hca e, as e idenced by hei g owing
use in p edic i e analy ics, pe sonalized medicine, and d ug
disco e y [6]. When in eg a ed wi h WBANs, AI/ML can
enable physicians o p ocess biosenso da a in eal ime,
c ea e pe sonalized ca e p og ams, and de ec heal h issues
ea ly [7].
WBANs, enhanced by ML, can pe o m c i ical asks
such as ac i i y ecogni ion o pe sonalized i ness plans
and anomaly de ec ion o iden i y unusual changes in i al
signs, po en ially p e en ing se e e heal h complica ions.
Fo ins ance, moni o ing hea a e a iabili y h ough ML
could assis in ea ly ca diac p oblem de ec ion, demon-
s a ing he ans o ma i e po en ial o hese sys ems. Fu -
he mo e, AI-d i en communica ion p o ocol op imiza ion
in WBANs imp o es da a ans e e iciency and ensu es
imely access o c i ical heal h in o ma ion. None heless,
he e a e s ill ongoing issues, especially in da a p i acy,
which equi es he p o ec ion o p i a e heal h in o ma-
ion, and ene gy e iciency, whe e ba e y li e is a limi ing
ac o . Ongoing esea ch add esses hese issues by de el-
oping p i acy-p ese ing models ha sa egua d pa ien da a
wi hou sac i icing pe o mance, as well as ene gy-e icien
algo i hms ha ex end de ice li espan. In addi ion, se e al
su eys ha e e iewed WBANs; mos ocus on IoT in eg a-
ion, ene gy-e icien ou ing, o speci ic ML applica ions
such as human ac i i y ecogni ion. O he s emphasize ene gy
op imiza ion, classi ica ion echniques, WBAN design, secu-
i y, o gene a i e AI in niche heal hca e asks. Howe e ,
no one o e s a comp ehensi e, ML- ocused syn hesis ha
links algo i hmic echniques di ec ly o WBAN heal hca e
challenges.
Along wi h imp o emen s o algo i hms, imp o e-
men s o wi eless communica ion ha dwa e and sensing,
like mic owa e-based senso s, subs a e-in eg a ed and
leaky-wa e an enna sys ems, will be e y impo an o in e-
g a ing ML in o WBANs o make physiological moni o ing
mo e accu a e and lexible [8],[9],[10],[11].
The cu en s udy ills hese gaps by conduc ing an SLR
on ad anced ML/AI applica ions in WBANs using he
P e e ed Repo ing I ems o Sys ema ic Re iew and Me a-
analysis (PRISMA) guidelines, wi h an emphasis on solu ions
o ene gy e iciency, quali y o se ice (QoS), secu i y,
anomaly de ec ion, ac i i y ecogni ion, communica ion, and
heal hca e applica ions. Unlike p e ious su eys, his e iew
ollows a s uc u ed, ep oducible me hodology o consol-
ida e agmen ed knowledge, classi y ML app oaches, and
iden i y pe sis en esea ch gaps by answe ing he ollow-
ing esea ch ques ion: ‘‘How can ML algo i hms be u ilized
in WBANs o sol e p oblems, imp o e pe o mance, and
di ec u u e ad ancemen s in heal hca e applica ions?’’ This
e iew’s main goals a e o summa ize he s a e-o - he-a in
ML-d i en WBANs o heal hca e applica ions, poin ou
he ad an ages and disad an ages o cu en s a egies o
ackling issues wi h ene gy, secu i y, and eal- ime da a anal-
ysis, and d aw a en ion o unexplo ed a eas while sugges ing
u u e lines o inqui y.
The ollowing a e he p ima y con ibu ions o his
s udy:
•The ou p ima y applica ion a eas o ML-enabled
WBANs a e me hodically iden i ied in his sys ema ic
li e a u e e iew (SLR). I in oduces a new axonomy
ha ames cu ing-edge me hods like gene a i e AI
and ede a ed lea ning (FL) wi hin he WBAN en i on-
men , while highligh ing ac i i y ecogni ion, anomaly
de ec ion, communica ion op imiza ion, and pe sonal-
ized ca e as he p ima y domains.
•The SLR pinpoin s a ious di ec ions, including ene gy
consump ion, along wi h da a-quali y cons ain s and
p i acy isks, and opaque model beha io as he main
obs acles o clinical adop ion. The s udy p oposes
speci ic s a egies, including ligh weigh model design
and explainable AI (XAI) echniques, which enhance
eal- ime sys em pe o mance while educing powe
usage and main aining pa ien p i acy.
•To acili a e ea ly disease de ec ion, ongoing i al sign
moni o ing, and in elligen esou ce managemen , all
o which con ibu e o a pa ien -cen ed app oach o
heal hca e, ou sugges ed a chi ec u al e iew in eg a es
well-known WBAN ha dwa e wi h deep ein o cemen
lea ning (DRL) elemen s and hyb id algo i hms.
•The compa a i e analysis demons a es ade-o s
be ween accu acy and la ency agains ene gy e iciency
and in e p e abili y while also iden i ying scalabili y
194730 VOLUME 13, 2025
A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
limi a ions in la ge-node WBAN deploymen s. The
assessmen de e mines which op imiza ion a eas equi e
immedia e a en ion.
•To highligh un esol ed issues, such as p i acy p o ec-
ion in high- olume senso s eams, WBAN in e e ence
managemen , edge-based ML p ocessing, and neu al
in e ace in eg a ion, he analysis syn hesizes indings
om a ious disciplines. I hen sugges s s a e-o - he-a
echnical solu ions wi h s ong da a secu i y and p i acy
sa egua ds ha adhe e o e hical and legal s anda ds.
This will help o guide u u e esea ch, build pa ien
us , and p omo e collabo a i e inno a ion.
Fo he eade ’s con enience, we ha e included a lis o
ac onyms used in Table 1. The emainde o his pape is
o ganized as ollows: Sec ion II p esen s he backg ound o
he s udy. Sec ion IV Rela ed Wo ks. Sec ion IV p o ides
a sys ema ic li e a u e e iew. In Sec ion V, ad anced ML
o WBANs is explained. Sec ion VI p esen s a discussion
and s a is ics o he indings. Sec ion VII discusses he chal-
lenges and u u e esea ch di ec ions o in eg a ing ML
in o WBANs, and conclusions and sugges ions o addi ional
esea ch a e p esen ed in Sec ion VIII.
II. BACKGROUND OF THE STUDY
Acco ding o a s udy by Abdu e al. in [12] WBANs a e
WSNs posi ioned as an eme gen echnological pa adigm ha
in ends o help heal hca e sys ems unc ion mo e e ec i ely.
WBANs a e in en ionally designed o suppo and allow
eal- ime heal h moni o ing and ocus on imely in e en ion
and ea men o people wi h li e- h ea ening condi ions [13].
Senso s in WBAN ope a e wi hin o ou side he human body
o collec a ious physiological signals and ans e hem o
a cen al node called he ‘‘coo dina o ’’ node [14]. WBANs
ha e a ac ed much a en ion because o hei po en ial appli-
ca ions in heal hca e, which enable hem o ou inely moni o
impo an physiological signs, ack pa ien mobili y, and
agg ega e da a ela ed o a ious medical diseases [15]. This
makes hem ideal o emo e pa ien moni o ing, all de ec-
ion, and ch onic disease managemen applica ions. A dis inc
end in schola ly discou se is a s ong emphasis on de el-
oping minia u ized, low-powe senso s and wea able de ices
ha will smoo hly i wi h he ope a ional amewo k o
WBANs [16]. Physiological da a such as blood p essu e,
glucose le els, body empe a u e, hea a e, and muscle
ac i i y could be measu ed by hese senso s [17]. Resea che s
ha e in es iga ed a ange o communica ion echnologies,
such as Blue oo h, Zigbee, UWB, and 5G. E e y echnique
has p os and cons, and he decision is usually in luenced
by he equi emen s o he applica ion [18]. WBAN de ices
mus ope a e on a limi ed amoun o ba e y powe . Powe
consump ion is an impo an conside a ion in inc easing he
li espan o WBAN de ices [19]. Many s udies ha e concen-
a ed on senso designs, powe managemen echniques, and
ene gy-e icien communica ion p o ocols [20].
TABLE 1. Commonly used abb e ia ion in he SLR.
The eme gence o wea able pla o ms such as sma -
wa ches and i ness acke s has boos ed WBAN g ow h [21].
VOLUME 13, 2025 194731
A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
TABLE 1. (Con inued.) Commonly used abb e ia ion in he SLR.
These consume -o ien ed de ices ha e he po en ial o
acqui e use ul heal h- ela ed da a and play a signi ican ole
in u u e heal hca e applica ions. The WBAN communica ion
a chi ec u e consis s o h ee p ima y unc ional componen s,
each o which is essen ial o enabling e icien da a ans e
and ne wo k connec ion [22],[23]. Ne e heless, Figu e 1
shows he de ails.
Tie -One:
Since hey se e as he basis o e ec i e da a in e change
and ne wo k in e ac ion, he communica i e dynamics in he
WBAN amewo k cons i u e a c ucial ocus o his pape .
This ie ocuses on he in e ac ion o biomedical senso
nodes, which measu e physiological pa ame e s, including
body empe a u e, hea a e, oxygen sa u a ion, and blood
p essu e, and send he ga he ed da a o a selec ed coo dina o
o sink node. WBANs ely on well-de ined wi eless s anda ds
o keep senso s alking while using as li le powe as possible.
The mos common choice is he IEEE 802.15.6 p o ocol, buil
speci ically o body-wo n de ices ha balance low ene gy
use wi h eliable links and quali y-o -se ice suppo [25].
Depending on he job, how a he signal mus a el,
how quickly da a needs o mo e, and how much ba e y li e
ma e s, designe s also u n o Blue oo h Low Ene gy, Zig-
bee, o LoRa. Tha choice shapes he ne wo k’s day- o-day
pe o mance, in luencing delays, ba e y d ain, and esis ance
o in e e ence.
A he link laye , he sys em manages da a agg ega ion, col-
lision a oidance, and channel access so ha e e y senso can
alk o he coo dina o wi hou hiccups. To push pe o mance
u he , enginee s add unc ions like coope a i e elaying,
dynamic channel selec ion, and adap i e modula ion and cod-
ing. These upg ades help he ne wo k s ay solid e en when
signals ade, pa ien s mo e a ound, o ou side de ices cause
noise.
Running ligh da a-p ocessing asks on he coo dina o ,
an edge-compu ing s ep, also eases he bu den on iny senso
nodes and cu s ansmission lag. Toge he , hese measu es
keep he WBAN scalable, compa ible wi h o he gea , and
quick enough o eal- ime moni o ing. They lay he g ound-
wo k o highe -le el asks such as da a analysis, anomaly
de ec ion, and au oma ed medical decisions.
Tie -Two:
The in e -WBAN laye handles e e y hing beyond he
coo dina o inside a single body ne wo k. I co e s body-
o-body (B2B) links, o e lapping WBANs, and connec ions
o ex e nal access poin s [26]. I s job is o enable mul iple
WBANs o communica e wi h one ano he and wi h ex e -
nal sys ems, cloud se e s, hospi al da abases, o emo e-
moni o ing dashboa ds, allowing da a o low smoo hly
whe e e i needs o go. Mos links a his le el u ilize amil-
ia wi eless echnologies, including Wi-Fi, cellula (4G/5G),
o long- ange, low-powe op ions such as LoRaWAN. The
choice depends on ange, bandwid h, delay, and powe use.
LoRaWAN, o ins ance, is g ea when you need kilome-
es o each bu ha e igh ba e y limi s, while 5G shines
in eal- ime scena ios ha demand millisecond la ency and
eliabili y [27].
Issues a e common in c owded a eas, busy hospi al
wa ds, and e ail malls. Signal clashes occu when nume -
ous WBANs use he same equency spec um. Sma
in e e ence-mi iga ion s a egies, dynamic spec um alloca-
ion, and equency hopping a e some o he echniques ha
keep he equency clea . The e a e addi ional obs acles o
B2B links: indi iduals mo e andomly, and bodies obs uc
signals. Coope a i e elays o adap i e ou ing p o ocols help
sides ep hose issues [28].
In e ope abili y ma e s jus as much as connec i i y.
S anda ds such as IEEE 802.15.6, along wi h he la es
5G and IoT specs, lay ou he ules o secu e, e icien
da a exchange. Ai igh secu i y, which includes end- o-end
enc yp ion, blockchain-based au hen ica ion, and p i acy-
p ese ing agg ega ion, is essen ial due o he sensi i e heal h
da a in ol ed.
By ackling in e e ence, ollowing open s anda ds, and
leaning on mode n wi eless ech, his laye unde pins a
obus , scalable WBAN ecosys em [29]. In sho , i is he
b idge ha links on-body ne wo ks o he wide heal hca e
wo ld.
Tie -Th ee:
The Beyond-WBAN laye allows o he ansmission o
in o ma ion o o he des ina ions, like heal hca e acili ies,
cloud s o age, o elemoni o ing sys ems. This laye es ab-
lishes a communica ion model ha anscends indi idual
WBAN bounda ies. In his laye ed a chi ec u e, he AP plays
an essen ial ole by allowing he passage o da a packe s om
he WBAN o wha e e des ina ion ia mul iple communi-
ca ion sys ems. I co e s all channels ha ensu e seamless
connec i i y and global each: sa elli e communica ions, cel-
lula ne wo ks, he In e ne , and WANs.
This ie should especially be conside ed by applica ions
such as elemedicine, emo e heal h moni o ing, and big da a
analy ics. Fo example, wea able senso s in a WBAN ansmi
194732 VOLUME 13, 2025
A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
FIGURE 1. WBAN a chi ec u e [24].
pa ien in o ma ion o cen alized heal hca e sys ems ha
in eg a e he da a wi h elec onic heal h eco ds (EHRs), s o e
i o he long e m, o analyze i in eal ime. By enabling
e ec i e da a p ocessing, s o age, and decision-making a
emo e loca ions, he use o cloud and edge compu ing
enhances his ie ’s capabili ies.
The Beyond-WBAN laye add esses one o he mos
impo an challenges: ensu ing eliable and secu e da a ans-
mission h ough ola ile o high-la ency communica ion
channels. Mul i-pa h ou ing, e o co ec ion mechanisms,
and QoS enhancemen solu ions play a i al ole in main-
aining da a in eg i y while minimizing la ency. Also, hese
echniques comp ess and agg ega e he da a o use less band-
wid h and ansmi mo e e icien ly, since many WBANs
gene a e massi e amoun s o in o ma ion [30].
Because in o ma ion sen ou side he WBAN pe ime e can
eadily a el h ough un us ed ne wo ks, inc easing he isk
o in e cep ion and unau ho ized access, secu i y and p i acy
a e c ucial issues.
Ad anced secu i y measu es such as blockchain-based
au hen ica ion, secu e unnelling p o ocols like TLS/SSL, and
end- o-end enc yp ion can indeed mi iga e hese h ea s [31].
Mo eo e , while suppo ing collabo a i e analy ics among
di e en WBANs, p i acy-p ese ing app oaches such as FL
and di e en ial p i acy ensu e ha con iden ial heal h in o -
ma ion emains p o ec ed.
S anda diza ion a his le el is also equi ed because in e -
ope abili y is wha allows seamless da a exchange be ween
di e en heal hca e sys ems and communica ion echnolo-
gies. In Beyond-WBAN en i onmen s, s anda ds such as
IEEE 802.15.6, IoT-speci ic s anda ds, and 5G-enabled
heal hca e amewo ks p o ide sa e and e ec i e commu-
nica ion guidance while also ensu ing be e collabo a ion
among hese sys ems. In eg a ing ML and AI a his le el
u he enables p edic i e analy ics, anomaly de ec ion, and
pe sonalized heal hca e ecommenda ions om he agg e-
ga ed da a o di e se sou ces.
The Beyond-WBAN ie c ea es a c ucial link o da a
dissemina ion ou side he immedia e WBAN a ea by ack-
ling hese issues and u ilizing inno a i e communica ion
and secu i y echnologies, as in [31]. This s a um enables
emo e diagnos ics and eal- ime heal h moni o ing and sup-
po s ex ensi e heal hca e p og ams like disease su eillance,
popula ion heal h managemen , and in e na ional heal h
esea ch.
A. AREAS FOR WBAN APPLICATION
WBAN has se e al p esen and po en ial uses as an ex en-
sion o WSN [32]. Widesp ead WBAN use in people’s daily
li es is made possible by he cu en inc ease in in e ne
speed, he global numbe o in e ne use s, he a o dabili y
o wea able echnology, and ad ancemen s in AI and big da a
analy ics.
WBAN con o ms o ne wo ked de ices (wi eless senso s)
communica ing ac oss sho dis ances wi hin o ou side he
body. WBAN has se e al medical applica ions, as depic ed
in Figu e 2.
1) MILITARY TRAINING AND SPORTS
Se e al dea h inciden s in spo s ha e been epo ed, and
medical esul s e ealed ha mos casual ies we e caused
by he ailu e o he body’s nume ous impo an o gans due
o a igue. Recen ly, wea able WBAN de ices ha e been
used o moni o he heal h condi ions o a hle es du ing
physical aining and mili a y soldie s in a ious mili a y
ope a ions [13].
VOLUME 13, 2025 194733

A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
FIGURE 2. WBAN applica ions.
2) RESCUE OPERATIONS AND DISASTER MANAGEMENT
Communica ion and o he in as uc u es o en ail o pe -
o m hei expec ed unc ions du ing a c isis. Rescue wo ke s
migh be unable o comple e a escue ope a ion i he
dependable communica ion ne wo k mal unc ions. WBAN
co e age dis ance is limi ed, excep o i s use in in o -
ma ion moni o ing (heal h, e c.). As a esul , he au ho s
ecommended using WBAN in addi ion o CR. Using he
unlicensed equency spec um ou side he ISM band, he
CR ansmi s da a based on oppo unis ic channel access,
connec ing da a in he WBAN ne wo k o he a ge ed heal h
uni s [33].
In sma homes, WBAN senso s (gy oscopes and
accele ome e s) a e used o ack he mo emen s and shi ing
pos u es o elde ly esiden s ecei ing assis ed li ing. People
can also easily connec o WBAN heal hca e sys ems using
emo e con ols o educe hei dependence [13]. When se i-
ous occu ences such as he de elopmen o a slump need o
be add essed, moni o ed ac ions ale o he amily membe s
and ca egi e s o in e ene.
3) CONSUMER ELECTRONICS AND ENTERTAINMENT
Millime e -wa e de ices enable e y sho - ange wi eless
links be ween gadge s. Wea ables such as sma wa ches and
i ness bands can connec o sma TVs, powe gaming con-
olle s, o s eam MP3s o e Blue oo h, demons a ing how
WBANs in eg a e in o oday’s en e ainmen and daily ou-
ines [34]. As echnology de elops, we wi ness inc easingly
cus omized wea ables and sma gadge s. Wea able echnol-
ogy has e ol ed beyond heal h o become a ashion s a emen ,
hanks o 5G’s highe speeds.
4) EMERGENCY HEALTHCARE SERVICES, SURVEILLANCE,
AND DEVICE MONITORING
Su eillance de ices, such as unmanned ae ial ehicles and
in eg a ed sa elli es, use high-de ini ion came as and o he
senso s o ack a ange o social ac i i ies, such as posi-
ion changes, in e ac ion, gai mo emen , and he ans e o
heal h da a o da a se e s ia UWB anscei e s. In addi ion
o su eillance, unmanned ae ial ehicles (UAVs) linked o
WBANs can be used o collec isual heal h da a om se e al
sou ces. Depending on he ype o da a, con olle s, doc o s,
and enginee s mus eac o cu en ci cums ances based on
he heal h s a us o people o equipmen [35]. In emo e o iso-
la ed places, WBAN pai ed wi h sa elli es can p o ide heal h
eme gency assis ance wi h minimal delays in physiological
signals.
In emo e de ices, embedded senso s (e.g., WBAN) mea-
su e a ious a ibu es du ing communica ion and coo dina-
ion. Remo e de ices can in e ac wi h he con ol s a ion,
sa e da a in he logge (o se e s), and independen ly
con ol hei ope a ions, depending on he embedded in el-
ligence [36].
5) HEALTH MONITORING
WBAN is a c ucial componen o heal h moni o ing. WBAN
senso s ha e been in eg a ed in o he human body as wea able
de ices, su ace con ac s, o implan s o collec and ansmi
i al pa ien heal h da a o emo e medical acili ies. Linked
biosenso s measu e nume ous pa ame e s, including blood
suga , hea a e, and body empe a u e. Acqui ed heal h da a
a e u ilized o p edic he cou se o a disease o o ecommend
emedial ac ions using AI algo i hms [37].
Mo eo e , WBAN has nume ous uses in s oke- ela ed
illnesses, diabe es managemen , ea ly cance cell de ec ion,
ehabili a ion p og ess acking, and c i ical ca e uni decision
assis ance [34],[38].
6) TELEMEDICINE
WBAN o e s ne wo k in as uc u e o elemedicine se -
ices, enabling pa ien s o ecei e i ual heal h ca e apidly,
in addi ion o moni o ing as onau s’ heal h du ing space
explo a ion [29]. Heal h s a us e alua ions and decisions a e
made by coo dina ing biological senso s, da a p ocessing
uni s, and minia u e ac ua o s. The ac ua o s hen pe o m he
decision as eedback. Fo ins ance, i i senses ele a ed blood
suga , he p ocesso ins uc s he ac ua o o egula e insulin
in he blood a e ies.
B. MACHINE LEARNING
Doc o s now ha e a new assis an in AI ha can help hem
wi h diagnosis and e en p ognosis. A ho ough unde s anding
o a pa ien ’s p oblems may also be acili a ed by he capaci y
o lea n om expe ience. WBANs ha employ AI algo-
i hms enable he c ea ion o compu e p og ams ha lea n
and de elop om expe ience, ins ead o being speci ically
designed o make p edic ions o ecommenda ions. O e he
194734 VOLUME 13, 2025
A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
las ew decades, ad ances in compu e powe ha e enabled
he de elopmen o esou ce-in ensi e AI echniques such as
ML solu ions. This s udy also illus a es how ML p og ams
s a is ically c ea e a p edic i e model using da a samples o
in o ma ion o aining. This aining is used o classi y
objec s o make p edic ions in applica ions necessa y o wise
decisions. Heal h da a collec ed om senso s is p ocessed
by sma heal hca e apps employing ML logic be o e being
sen o he cloud o p ocessing by ML algo i hms. The
esul s we e su icien ly eplica ed, and he collec ed da a
we e e e ed o as he es ing da a. The ou comes will also
be u ilized in he aining phase o he subsequen es ing da a
a e hey ha e been eplica ed. As a esul , he da a collec ed
by he senso s is conside ed es ing da a and, a e p ocess-
ing, aining da a o addi ional medical e alua ions. Robus
implemen a ion, de ec ion accu acy, and compu a ional cos
a e c ucial when selec ing ML algo i hms o WBAN
applica ions [39].
DL, supe ised lea ning, and unsupe ised lea ning a e he
main ca ego ies in which ML sys ems a e inco po a ed wi h
WBAN applica ions. Figu e 3illus a es how ML p og ams
s a is ically c ea e a p edic i e model using da a samples o
in o ma ion o aining.
FIGURE 3. Machine lea ning p ocess [39].
1) SUPERVISED LEARNING
Supe ised lea ning is he p ocess o aining algo i hms o
classi y o p edic da a o ou comes co ec ly using labelled
da ase s. The objec i e is o each he algo i hm a gene al ule
ha links inpu s o ou pu s by p o iding ins ances o inpu s
and he expec ed ou pu s ha hey should gene a e. WBANs
use supe ised lea ning algo i hms, especially whe e da a can
be app op ia ely ca ego ized and a weal h o p io knowledge
abou he human body is a ailable. One p ominen example is
he es ima ion o a mobile node’s loca ion using an algo i hm
ained on signal p opaga ion cha ac e is ics (inpu s) and
designa ed loca ions (ou pu s). The applica ion o supe ised
lea ning has e ec i ely handled se e al obs acles wi hin he
ield o WBANs, including he MAC [40], ou ing [41],
disease p edic ion in WBAN [42], de ec ion o human body
mo emen s in WBANs [43], human ac i i y ecogni ion [44],
and secu i y [45],[46].
2) UNSUPERVISED LEARNING
Unsupe ised lea ning e e s o algo i hms ha ind pa -
e ns in da ase s, including unlabelled and unca ego ized
da a poin s. The lea ning algo i hm ac s as a ea u e ex ac-
ion ool, unco e ing hidden pa e ns in he da a wi hou
labelling. P oblems, such as au oma ically clus e ing wi eless
senso nodes acco ding o hei cu en obse ed da a alues
(wi hou knowing he g oup membe ship o each node be o e-
hand), can be sol ed wi h unsupe ised lea ning.
Anomaly de ec ion, such as clus e ing o ou lie de ec ion,
can be used o iden i y unexpec ed pa e ns o abno mali ies
in physiological da a acqui ed by WBAN senso s [47],[48].
This is pa icula ly use ul in heal hca e applica ions, whe e
de ia ions om no mal pa e ns may indica e heal h issues
o senso mal unc ions. Da a comp ession and dimensional-
i y educ ion can be achie ed using unsupe ised lea ning
me hods, such as PCA. WBANs bene i om his solu-
ion because i enables he ansmission o i al in o ma ion
while minimizing da a ans e . Pa ien p o iling h ough
clus e ing algo i hms enables he g ouping o pa ien s based
on simila heal h cha ac e is ics d awn om WBAN sen-
so in o ma ion [49]. This can help c ea e de ailed pa ien
p o iles, making i possible o o e pe sonalized heal hca e
ea men s and app oaches [50],[51]. I also suppo s ne -
wo k sel -o ganiza ion, such as enabling WBAN nodes o
g oup hemsel es based on sha ed cha ac e is ics and o m
clus e s [52]. This sel -o ganiza ion makes ne wo k commu-
nica ion mo e e icien o e all by allowing o be e esou ce
alloca ion and coo dina ion, as well as dynamic channel
alloca ion, which means ha channels can be analyzed o
condi ions and in e e ence pa e ns wi hou supe ision.
This in o ma ion is hen u ilized o dynamically assign chan-
nels, dec ease collisions, and imp o e he dependabili y o
da a ansmission wi hin a WBAN [53].
3) REINFORCEMENT LEARNING
RL is a p ominen ad anced sub ield o ML [54]. He e,
an agen is assigned he esponsibili y o making a sequence
o decisions while in e ac ing dynamically wi h i s en i-
onmen . The basic objec i e o RL lies in imp o ing he
decision-making policies o accumula e a ewa d signal,
which is a ained h ough i e a i e explo a ion based on ial
and e o . This p ocess is o malized using he amewo k
o a Ma ko Decision P ocess (MDP), which o e s a s uc-
u ed ma hema ical model o p oblems in ol ing sequen ial
decision-making unde unce ain y. RL, along wi h supe -
ised and unsupe ised lea ning, is one o he h ee co e ML
pa adigms. RL algo i hms can sense and unde s and hei
su oundings, ac and lea n om ailu es, and choose he
bes ac ion o in elligen agen s o maximize he cumula i e
ewa d in each en i onmen . The lea ning algo i hm in e ac s
wi h a dynamic en i onmen as i mo es h ough i s p oblem
a ea and ecei es eedback ha i may compa e o ewa ds,
which i aims o op imize [39]
VOLUME 13, 2025 194735
A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
The esea ch ha depic s one ype o RL echnique,
which is he Q-lea ning p ocedu e, is used o slo dis i-
bu ion in a WBAN se ing. This illus a es how Q-lea ning
can be applied o in elligen and adap i e esou ce alloca-
ion in WBANs by highligh ing he in e ac ions among he
agen , en i onmen , s a e ansi ions, and incen i es. The
amewo k’s mul iple componen s ope a e in a ein o cemen
lea ning loop o maximize esou ce alloca ion [55].
An agen ep esen ing one o he WBAN nodes begins he
p ocess by in e ac ing wi h i s su oundings. The agen acks
he WBAN nodes’ cu en s a e (S ), which encompasses se -
e al ne wo k s a es and me ics. Based on his s a e, he agen
chooses an ac ion (a ), like how o assign communica ion
slo s, o keep e e y hing unning smoo hly.
In he WBAN en i onmen , nodes communica e wi h each
o he o ca y ou an ac ion and see wha happens. I he ac ion
wo ks well, he en i onmen mo es o a new s a e (S +1), and
he agen ge s a ewa d (R ). This ewa d shows how well
he chosen slo alloca ion helps imp o e sys em pe o mance,
like educing delays o sa ing ene gy.
The Q-lea ning algo i hm helps he agen imp o e i s
decision-making based on he ewa ds i ecei es and he
new si ua ions i encoun e s. Th ough epea ed lea ning, he
agen g adually igu es ou he bes way o alloca e ime slo s.
I does his by balancing he need o y new ac ions (explo-
a ion) using wha i al eady knows wo ks well (exploi a ion),
allowing i o adjus o changes in he ne wo k. This adap abil-
i y helps imp o e he pe o mance o he WBAN sys em.
Figu e 4shows how Q-lea ning enables sma and lexible
esou ce managemen in WBANs by illus a ing he in e ac-
ions be ween he agen , he en i onmen , s a e changes, and
ewa d signals.
FIGURE 4. Rein o cemen lea ning.
4) DEEP LEARNING
DL is a subse o supe ised lea ning echniques whe e ou pu
is gene a ed by p ocessing inpu h ough some nonlinea
ans o ma ions. Re e ence [56] claims ha DL makes i
possible o compu a ional models composed o mul iple p o-
cessing laye s o lea n da a ep esen a ions a di e en le els
o abs ac ion. The capaci y o DL o au oma ically ex ac
high-le el ea u es om complex da a is a key ad an age
o e adi ional ML sys ems. Handc a ing p e ious ea u es
is much simpli ied by he ac ha he lea ning p ocess does
no ha e o be cons uc ed by a human [56].
Howe e , he in e p e abili y o he model su e s due o he
pe o mance o DL echniques, such as deep neu al ne wo ks
(DNNs). Since DNNs make unique decisions, hey a e some-
imes pe cei ed as ‘‘black boxes’’ wi h li le unde s anding.
Addi ionally, DNNs o en ha e nume ous hype pa ame e
uning issues, making i di icul and ime-consuming o
de e mine he bes con igu a ion. T aining DL ne wo ks can
also be compu a ionally expensi e, equi ing obus pa -
allel compu ing capabili ies like g aphics p ocessing uni s
(GPUs). The e o e, i is c ucial o conside he ene gy and
compu ing limi a ions o embedded o mobile de ices when
deploying DL models. Fu he mo e, DL can lea n o pe o m
ca ego iza ion asks di ec ly om ex , audio, o images. Syn-
he ic neu al ne wo ks (NNs) suppo hese unc ions. Senso s
collec i al sign da a, allowing AI o iden i y pa e ns
ins an ly. AI-powe ed iage sys ems expedi e eme gency
esponse by p io i izing cases based on hei u gency. La ge
olumes o da a gene a ed by biosenso s a e eadily and
swi ly p ocessed wi h he help o AI app oaches. This quick
p ocessing educes la ency by acili a ing imely diagnosis.
AI-enabled emo e moni o ing de ices, au oma ed pa ien
sc eening, eal- ime ale s, and p edic i e algo i hms also
help physicians make in o med decisions and signi ican ly
educe eac ion imes. NN a chi ec u es wi h mul iple laye s
and subs an ial amoun s o sample da a a e used o de elop
DL models o each s a e-o - he-a accu acy [57]. Two exam-
ples o analy ical DL models ha p o ide compu a ional
in elligence solu ions by s udying as da ase s a e CNN and
deep belie ne wo ks, which a e used when shallow lea ning
canno assess he necessa y meaning ulness o ends. P eci-
sion medicine o en uses hese lea ning models o diagnose
illnesses and gene a e new ea men s [58].
C. MACHINE LEARNING ALGORITHMS
This sec ion akes a close look a he ML app oaches com-
monly used in WBAN esea ch, wi h a c i ical e iew o hei
s eng hs and limi a ions.
1) SUPPORT VECTOR MACHINE
SVM is a me hod used o sol e classi ica ion p oblems.
I wo ks by ans o ming he da a in o a highe -dimensional
space, whe e a hype plane can sepa a e he di e en classes
clea ly.
Figu e 5illus a es he ou -laye ed a chi ec u e, show-
ing how WBANs can be in eg a ed wi h ML algo i hms o
enhance heal hca e se ices.
194736 VOLUME 13, 2025
A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
FIGURE 5. Fou -laye ed a chi ec u e illus a ing how WBANs can be combined wi h ML algo i hms o imp o e heal hca e se ices [59].
The BAN laye uses biosenso s h oughou he body o
ack c i ical indica ions, including empe a u e, oxygen sa -
u a ion, and hea a e. These senso s con inuously collec
physiological da a. Howe e , he use in e ace laye allows
cellula ne wo ks o a wi eless ou e o ga he da a wi elessly
and send i o consume de ices, including sma phones and
able s (such as he iPhone). This laye makes i easie o
e ie e da a ini ially and connec o he in e ne o addi-
ional p ocessing. In addi ion, he in o ma ion ha is sen
o a se e ha applies supe ised ML echniques, including
andom o es (RF) and decision ee (DT), and SVM s ages,
including da a collec ing, il e ing ( o elimina e noise o
unnecessa y in o ma ion), and analysis, makes up his laye .
A e analysing he il e ed da a, he algo i hms p oduce use-
ul insigh s o heal hca e decision-making. Las ly, doc o s
and o he medical p o essionals can u ilize he ML analysis’s
conclusions o moni o ing, diagnosis, o eme gencies. Doc-
o s can e alua e pa ien s’ heal h and adminis e ea men s
om a dis ance. Eme gency se ices a e no i ied o a end
u gen medical demands in c i ical ci cums ances.
2) RANDOM FOREST
One ype o supe ised ML me hod ha wo ks well o bo h
eg ession and classi ica ion p oblems is RF. I alls unde
ensemble lea ning s a egies, especially he bagging (boo -
s ap agg ega ing) echnique.
A bagged decision ee-based ensemble lea ning echnique
is called RF. Boo s ap agg ega ing, o bagging, is aining
se e al classi ie s on a ious subse s o he aining da a
and a e aging hei esul s o gene a e he inal p edic ion.
By success ully lowe ing he ensemble model’s a iance,
his me hod imp o es he model’s s abili y and p edic ion
p ecision. DTs, sensi i e o inpu da a changes and p one
o high a iance and o e i ing, a e especially well-sui ed
o bagging. RF is an e ec i e ool o classi ica ion and
eg ession asks because i combines he ou pu s o se e al
decision ees o p oduce s ong and dependable p edic ions.
Figu e 5 ep esen s a ou -laye ed a chi ec u e illus a ing
how WBANs, combined wi h SVM, RF, and DT algo i hms,
imp o e QoS.
3) K-MEANS
To a ange da a in o mu ually exclusi e g oups (o clus e s),
K-means aims o make obse a ions om he same clus e as
simila as possible while keeping obse a ions om di e en
clus e s as di e se as possible. In K-means clus e ing, each
clus e ’s cen e , o cen oid, co esponds o he means o he
obse a ion alues assigned o he clus e . Assigning se e al
poin s o he numbe o g oups o clus e s is in ended o p o-
duce high in a-clus e and low in e -clus e simila i ies [60].
Howe e , he da a ha is p oduced in WBAN is di e se.
Addi ionally, he dis ibu ion o biosenso s is no consis en .
The s udy demons a es ha he e ec i eness and depend-
abili y o da a collec ion and communica ion in WBANs
a e signi ican ly impac ed by da a clus e ing using an unsu-
pe ised ML echnique like K-Means [61]. The clus e ing
VOLUME 13, 2025 194737
A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
TABLE 2. (Con inued.) Compa ison o ecen and ela ed su eys on he WBAN applica ion.
as well as wi eless echnologies and a chi ec u al amewo ks
like LPWAN, 5G, and 6G, was he ocus o ea ly wo ks [13],
[97]. These con ibu ions p o ided ho ough mappings o
he echnologies and sys em a chi ec u es ha we e a ail-
able, bu hey equen ly lacked iewpoin s on ML and AI,
which limi ed hei abili y o mee he inc easing demand o
da a-d i en and adap i e WBAN solu ions.
By 2022, su eys had become mo e specialized, wi h
se e al s udies concen a ing on da a classi ica ion and
ene gy-e icien ou ing [98],[99],[100],[101]. These s ud-
ies acknowledged he po en ial o ML o enhance he
QoS and op imize ou ing. Ne e heless, hey p o ided li -
le ad ice on how o implemen ML in WBANs, e en
hough hey acknowledged i as a p omising ool. Gi en he
esou ce-cons ained na u e o WBAN de ices, whe e he
di ec applica ion o adi ional ML models is limi ed by
compu a ional complexi y, ene gy cons ain s, and ha dwa e
in eg a ion issues, such ad ice is especially c ucial. Appli-
ca ions o HAR and heal hca e we e in es iga ed in pa al-
lel [102],[103], whe e ML showed dis inc ad an ages in
classi ica ion and adap i e lea ning. Howe e , p oblems like
in e p e abili y, concep d i , and in eg a ion wi h eal- ime
communica ion channels we e s ill unsol ed.
Conside ing he g owing signi icance o WBANs in del-
ica e heal hca e se ings, mo e ecen su eys (2023–2024)
placed a g ea e emphasis on secu i y and p i acy. Th ea s
o da a in eg i y and con iden iali y we e s udied in s udies
like [6],[104],[105], and [106]. Some e en in oduced
gene a i e AI models o imp o e WBAN secu i y [79].
Domain-speci ic applica ions such as medical imaging o
he diagnosis o lung cance also demons a ed he expanding
ole o deep lea ning in WBAN-enabled heal hca e [107].
Despi e hese de elopmen s, mos su eys emain p ima ily
concep ual in na u e and ail o p o ide p ac ical alida ion.
Fu he mo e, e en hough ML/AI iewpoin s a e being in o-
duced g adually, in e p e abili y, compu a ional iabili y, and
p i acy-p ese ing s a egies like ligh weigh c yp og aphic
models o FL a e equen ly o e looked in discussions.
Conside ing his, ou su ey’s con ibu ion is i s ho ough
in eg a ion o ML/AI iewpoin s om all hema ic a eas o
WBAN esea ch, such as a chi ec u e, ou ing, QoS, ene gy
e iciency, p i acy, and heal hca e applica ions. This su ey
sheds ligh on he speci ic obs acles p e en ing ML adop ion
in WBANs, in con as o p e ious wo ks ha ei he igno e AI
conside a ions en i ely o concen a e only on a limi ed num-
be o echnologies [4]. These include he in e p e abili y o
p edic i e models, p i acy issues, da a quali y and a iabili y,
and ene gy and esou ce limi a ions. Ou su ey aids in he
c ea ion o in elligen , sa e, and e ec i e heal hca e appli-
ca ions by b idging communica ion-cen ic and AI-d i en
iewpoin s. I also o e s a comp ehensi e oadmap o u -
he ing WBAN esea ch. Table 2summa izes he de ails.
IV. SYSTEMATIC LITERATURE REVIEW METHODOLOGY
FOR ML IN WBANs
This s udy’s me hodology ollows he PRISMA and Me a-
Analyses amewo k, which ensu es ha he sys ema ic
e iew can be conduc ed in an open and epea able man-
ne [108]. Iden i ica ion, sc eening, eligibili y, and inclusion
a e he ou p ima y s eps in he PRISMA amewo k.
The iden i ica ion s ep in ol ed a comp ehensi e sea ch
using p e-se keywo ds ela ed o ad anced ML echniques
in WBANs in epu able da abases such as Web o Sci-
ence, Scopus, and IEEE Xplo e. In addi ion o e ms like
‘‘Machine Lea ning,’’ ‘‘Deep Lea ning,’’ and ‘‘Rein o ce-
men Lea ning,’’ he sea ch que y included ‘‘Wi eless Body
A ea Ne wo ks’’ o ‘‘WBAN.’’ A ocus on ML applica ions,
ele ance o WBANs, and publica ion in pee - e iewed jou -
nals we e among he inclusion and exclusion c i e ia used o
il e s udies du ing he sc eening phase. We excluded s udies
ha didn’ i hese equi emen s, such as no being in English
o being sho e han ou pages. Du ing he eligibili y p o-
cess, we used es ablished e alua ion measu es o ensu e
he alidi y and eliabili y o he s udies we selec ed. Da a
om quali ying s udies we e inally combined du ing he
inclusion phase o de e mine ends, obs acles, and po en ial
a enues o u u e esea ch ega ding he applica ion o ML
echniques o WBANs. The PRISMA amewo k is demon-
s a ed in he pape ‘‘Ene gy E icien and Reliable Rou ing
in Wi eless Body A ea Ne wo ks Based on Rein o cemen
Lea ning and Fuzzy Logic’’ by Guo e al. in [109]. Because
o i s subs an ial con ibu ion o imp o ing WBANs’ ene gy
194744 VOLUME 13, 2025

A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
FIGURE 9. PRISMA F amewo k [110].
TABLE 3. Lis o esea ch ques ions.
TABLE 4. Sea ch s ings used in he selec ed da abases.
e iciency and dependabili y, his wo k was ound du ing he
ini ial sea ch, alida ed o ele ance, e alua ed o quali y,
and inally included in he e iew. The PRISMA amewo k,
which ensu es a igo ous, me hodical, and objec i e e iew
p ocess, is used in his s udy ollowing bes p ac ices o
sys ema ic li e a u e e iews.
All ou o he SLR wo k low’s p ima y s eps a e depic ed
in Figu e 9.
A. IDENTIFICATION
1) FORMULATION OF RESEARCH QUESTIONS (RQS)
In de eloping he esea ch opic, wo sou ces we e used:
i s , concep s om p io s udies [103],[111]. All he pub-
lica ions ocused on how ML algo i hms can be bes used
wi hin WBANs o add ess di icul ies, imp o e pe o mance,
and lead o u u e imp o emen s in heal hca e applica ions.
Second, he mnemonic PICo, which s ands o ‘P’ (popu-
la ion o p oblem), ‘I’ (in e es ), and ‘Co’ (con ex ), based
on hese no ions, con ained h ee no ewo hy ea u es as pa
o he e iew [112]. The popula ion comp ises academics
and esea ch communi ies ocused on WBANs and ML. The
aim is o comple ely e iew and unde s and he use o ML
algo i hms in he con ex o WBANs and in es iga e a ious
ML echniques and hei impac on WBAN pe o mance in
heal hca e. The in e es lies in comp ehensi ely e iewing
and unde s anding he u iliza ion o ML algo i hms in he
con ex o WBANs and explo ing a ious ML echniques
and hei impac on he pe o mance o WBANs in heal h-
ca e. The se ing is WBAN- ela ed, wi h a ocus on ML
in eg a ion and applica ions. This enabled he au ho s o o -
mula e he h ee esea ch ques ions o his s udy as lis ed
in Table 3.
2) KEYWORDS FORMULATION
Six p ima y keywo ds we e ound based on he de eloped
s udy ques ions: ML echniques, AI, DL, RL, and he WBAN
app oach. To supplemen hese keywo ds, he au ho used an
in e ne hesau us such as hesau us.com, consul ed p e ious
s udies’ keywo ds, consul ed Scopus’ keywo ds, and sough
expe ad ice. Se e al e ms we e sea ched using his p ocess,
including ML echniques, AI, DL, RL, WBAN, and wea able
senso s.
VOLUME 13, 2025 194745
A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
3) SEARCHING CRITERIA USING DATABASES
Using ield code unc ions, ph ase sea ches, and Boolean
ope a o s, he keywo d combina ions we e looked up in
h ee da abases: Web o Science, Scopus, and IEEE Xplo e,
as shown in Table 4.
4) RECORD EXTRACTION
Whi emo e e al. [113] s a e ha using quali a i e o mixed
me hod app oaches ha enable he esea che o compa e
p ima y da a sou ces i e a i ely is he bes way o syn he-
size o analyse in eg a i e da a. In his s udy, an in eg a i e
e iew was used. Se e al pape s in ended o quan i a i e,
quali a i e, and combina ion e iews we e included hanks o
his p ocess. The esea che ca e ully examined he abs ac ,
esul s, and discussion po ions o he a icles. The da a
abs ac ion was based on he esea ch ques ions, meaning ha
all in o ma ion om he examined s udies ha could help
answe he esea ch ques ions was ex ac ed and included in
Table 5. Subsequen ly, he esea che employed a hema ic
analysis o disce n hemes and sub- hemes by examining he
abs ac ed da a o pa e ns and opics, clus e ing, numbe ing,
and iden i ying pa allels and co ela ions [114]. Fo syn he-
sizing a mixed s udy design (in eg a i e), hema ic analysis is
hough o be he mos sui able me hod [115]. I is cha ac e -
ized as a desc ip i e app oach ha connec s wi h o he da a
analysis app oaches and lexibly analyzes da a [116].
C ea ing hemes is he i s s age in a hema ic anal-
ysis. We looked o ends in he abs ac ed da a om
e e y e iewed pape h oughou he p ocess. Th ee majo
ca ego ies we e o med by pooling ele an o compa a-
ble abs ac ed da a in o a g oup. A e ha , he au ho s
e-examined he h ee da a g oups and disco e ed he emain-
ing hi y- wo subg oups. A e iew o hese hemes’ co -
ec ness was he nex s ep. The w i e s e-examined each
majo heme and sub- heme p oduced du ing his p ocess
o gua an ee hei applicabili y and co ec depic ions o he
da a. The w i e s hen mo ed on o he ollowing phase:
iden i ying he opics o e e y g oup and i s subg oup.
Be o e naming he opics o he subg oup, he au ho s i s
iden i ied he hemes o he leading g oup. This me hod
was used o es ablish hemes in a g oup o co-au ho s and
co esponding au ho s who sha ed he indings’ heme. Un il
he poin o ag eemen on modi ying he p oduced hemes and
sub- hemes, he esea che alked abou any con adic ions,
concep s, conund ums, o hough s ha migh be connec ed o
he in e p e a ion o he da a. Two panels o expe s, each wi h
backg ounds in communi y de elopmen s udies and qual-
i a i e me hodology, we e shown he p oduced hemes and
sub- hemes. The expe s we e asked o assess 32 sub- hemes
such as: Pe o mance e alua ion o he model, Ene gy-
E icien , Reliable, and Low-La ency, Reduc ion in Ene gy
Consump ion and La ency, Compu a ional Time, P ecision,
O e all Accu acy, Sensi i i y and Speci ici y, Th oughpu ,
Model Pe o mance Imp o emen , Hype pa ame e Tuning,
Model Ou pe o mance, Add essing Challenges, Reduc ion
in T ansmission Cos , E iciency and Reliabili y Enhance-
men , Imp o ed Sys em Ene gy E iciency, Inc eased U ili y
o WBAN, Inc eased E ec i eness o he Senso , Signi i-
can Inc ease in Sys em Li e ime, Powe Op imiza ion and
In e e ence Reduc ion. Fa ou able an i-in e e ence pe -
o mance and equency esou ce u iliza ion, Du y Cycle
Op imiza ion, Dec eased da a a ic, Success ul Da a Deli -
e y, Reduced A ack Possibili y o Jamme , Imp o ed Block
E o Ra e, QoS Sa is ac ion, Con ibu ions o Heal hca e
Decision-Making and Ou comes, Adap i e Pe o mance in
a Fas -changing WBAN Topology, High P edic ion Ra e o
Hea Diseases and Model E icacy in WBAN, De elopmen
o an ML-Based mHeal h Sys em, E ec i eness o Image
Enc yp ion/Comp ession, Con ibu ion o Reducing Misdi-
agnosis and False Posi i es, and 3 main hemes such as (1)
he mos common ML models u ilized in WBAN esea ch, (2)
pe o mance e alua ion me ics and he insigh s hey o e ,
and (3) u u e esea ch di ec ions and challenges. subjec-
i ely. Bo h concu ed ha he hemes and sub- hemes we e
app op ia e and pe inen o he e iew’s indings.
B. SCREENING
The second p ocess was sc eening, in which a icles we e
ei he added o o emo ed om he s udy depending on a
p ede e mined se o c i e ia (ei he manually by he au ho
o wi h he help o he da abase). This e iew es ic ed he
sc eening me hod o include publica ions published be ween
2017 and 2025 o adhe e o he concep o esea ch ield
ma u i y, which is highligh ed by [117]. This ch onology
was selec ed because he e was enough published esea ch o
conduc a ep esen a i e e iew. Because empi ical esea ch
pape s include p ima y da a, he w i e s chose o e iew hem.
To p e en misunde s andings, only English-language ex s
we e conside ed. As a esul , 2,407 a icles we e le o
e iew in he nex ound. Table 5depic s he de ails.
C. ELIGIBILITY AND INCLUDED
This sec ion combined he wo phases oge he , making i
an eligibili y and inclusion phase. In his sec ion, a o al,
633 documen s we e ob ained om he Web o Science, 1,123
om Scopus, and 651 om he IEEE Xplo e. Da a e ie ed
om sea ch engines included au ho s, names, i les, digi al
objec iden i ie s (DOIs), abs ac s, and keywo ds. The h ee
lis s we e hen in eg a ed in o Mendeley, emo ing edundan
a icles. Duplica e pape s we e ound and elimina ed using
he Mendeley e e ence manage o p ese e he in eg i y o
he eco ds ga he ed du ing he SLR. The da a was expo ed
in esea ch in o ma ion sys em (RIS) and comma-sepa a ed
alues (CSV) o ma s om academic da abases, including
Web o Science, Scopus, and IEEE Xplo e, and hen impo ed
in o Mendeley o addi ional p ocessing. Mendeley’s buil -in
Check o duplica es unc ion was used o compa e me ada a
ields, such as i les, au ho s, publica ion yea s, and DOIs,
o ind he duplica ed pape s sys ema ically.
194746 VOLUME 13, 2025
A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
TABLE 5. Inclusion and exclusion c i e ia.
The deduplica ion p ocedu e began wi h a e iew o he
duplica es ha we e ound. Mendeley au oma ically g ouped
a icles wi h me ada a ha ma ched o almos ma ched in o
se s o e i ica ion. A icles wi h simila names o iden i-
cal DOIs we e ho oughly inspec ed o con i m duplica ion.
Based on i le simila i ies, Mendeley’s ma ching sys em iden-
i ied po en ial duplica ion o en ies lacking DOIs. This was
b ough on by me ada a issues ha led o sligh a ia ions in
i les o au ho o de , which we e ixed by manual inspec ion.
A e his p ocedu e, he eco ds we e educed om 2,407
o 1,744 unique en ies a e 663 duplica e eco ds we e
de ec ed and emo ed. Unique a icles we e kep o e iew
once he cleaned da ase was expo ed o u he analysis.
By a oiding he inclusion o duplica e a icles in he li e a u e
e iew, ou Mendeley app oach expedi ed he deduplica ion
p ocess wi hou sac i icing accu acy. Figu e 10 depic s he
Mendeley deduplica ion p ocedu e.
Howe e , a o al o 1,501 ou -o -scope eco ds we e
emo ed om he 1,744 eco ds ob ained in Mendeley, and all
abs ac s we e e iewed o exclude ex aneous i ems om he
s udy’s objec i es. Ne e heless, he de ails o he da a ex ac-
ion ha e been iden i ied. In his s age, he da a ex ac ed o
inclusion and exclusion c i e ia is 243 a icles. Ou o he
numbe o selec ed a icles, 81 we e excluded based on i le,
162 we e selec ed based on hei i le and ele ance o he
s udy, and 14 epo s we e no e ie ed based on he abs ac .
Fu he mo e, 148 Repo s we e assessed o eligibili y, and
52 a icles we e excluded because hey did no ha e an ML
applica ion. 21 a icles we e emo ed o no being o iginal
con ibu ions, 9 o ha ing ewe han ou pages, 10 o no
being published in pee e iew, 3 o no being published
in English, and 3 o being e iew s udies. Finally, a e all
he assessmen s based on he exclusion and inclusion c i e-
ia, 55 pape s we e selec ed o he s udy [108]. Figu e 11
p esen s he PRISMA lowcha designed o his s udy.
D. BACKGROUND OF THE SELECTED STUDIES
Mo eo e , conce ning yea s o publica ion, 2 a icles we e
published in 2017–2018, 9 a icles we e published in
2019–2020, 20 a icles we e published in 2021–2022, and
24 a icles we e published in 2023–2025.
Figu e 12 shows he equency o a icles published
be ween 2017 and 2025. This indica es a no able inc ease
in he use o ML echniques in WBANs. The mos ecen
publica ion yea s a e 2023–2025, wi h 24 a icles.
E. SUMMARY OF THE SECTION
An ex ensi e e iew o ML me hods used in WBANs is
discussed in his sec ion. F om adi ional supe ised and
unsupe ised echniques o ad anced DL models, i ho -
oughly examines a a ie y o ML algo i hms, emphasizing
hei applica ions in ac i i y ecogni ion, anomaly de ec-
ion, communica ion op imiza ion, ou ing p o ocols, QoS,
pe sonalized heal hca e, and secu i y imp o emen s. The
popula i y and e ec i eness o ML models in ackling
WBAN issues, including ene gy e iciency, eal- ime moni-
o ing, and da a eliabili y, a e demons a ed by examining
55 chosen publica ions. RL maximizes ene gy usage and
adap abili y, while DL models, such as CNNs and LSTMs,
pe o m be e when handling complica ed da a. Ne e he-
less, he e a e s ill in e p e abili y, scalabili y, and p ocessing
cos issues. The axonomy o ML app oaches in WBAN
highligh s he signi icance o choosing he igh models o
ce ain applica ions and he need o mo e s udy o ge beyond
cu en obs acles and inc ease hei use ulness in heal hca e
sys ems.
V. BASIC AND ADVANCED ML-BASED TECHNIQUES FOR
WBANs
The impo ance o iden i ying equen ly used ML algo i hms
and unde s anding hei unc ions in handling he complex
and dynamic da a gene a ed by WBANs was unde lined by
his sec ion, in which a a ie y o app oaches ha e been s ud-
ied, including mo e complex DL models as pa o supe ised
lea ning s a egies [12],[111],[118],[119],[120],[121],
[122], o he mo e sophis ica ed DL models [63],[120],
[121],[123],[124],[125],[126],[127],[128],[129],[130],
[131],[132],[133],[134],[135],[136] and con en ional
unsupe ised echniques [137],[138]. The ad an ages o each
kind o algo i hm inc ease he p ecision, e ec i eness, and
adap abili y o WBAN applica ions. Because hey a e simple
o use and comp ehend, basic ML echniques like DT, SVMs,
and KNN a e well-liked in esou ce-cons ained en i on-
men s. Con e sely, sophis ica ed machine lea ning me hods
like CNNs, RNNs, LSTM ne wo ks, and RL excel a han-
dling high-dimensional da a, iden i ying in ica e ea u es,
and s eamlining dynamic p ocesses like slo alloca ion and
ene gy managemen . The 55 selec ed s udies demons a e ha
a wide a ie y o ML models a e being applied and a e e ec-
i e in WBANs [12],[63],[109],[111],[118],[119],[120],
[121],[122],[123],[124],[125],[126],[127],[128],[129],
VOLUME 13, 2025 194747
A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
FIGURE 10. Mendeley deduplica ion p ocess.
[130],[131],[132],[133],[134],[135],[136],[137],[138],
[139],[140],[141],[143],[145],[146],[147],[148],[149],
[150],[151],[152],[153],[154],[155],[156],[159],[163],
[164],[166],[167],[173],[174],[175],[176],[181],[182],
[183],[184]. Mos s udies clea ly iden i y he algo i hms
hey used. These indings demons a e how c ucial ML is o
encou age c ea i i y and imp o e WBAN pe o mance.
A. ML-BASED TECHNIQUES FOR WBANs
This sec ion p o ides a ho ough examina ion o he
ML-based me hods used in WBANs. To add ess c i ical
issues in WBANs, including ac i i y ecogni ion, anomaly
de ec ion, communica ion op imiza ion, ou ing p o ocols,
QoS, pe sonalized heal hca e, and secu i y enhancemen s,
i sys ema ically ca ego izes and e alua es a a ie y o ML
echniques. This e iew lines up a wide ange o ML
app oaches, om s aple supe ised and unsupe ised me h-
ods o newe DL, RL models, and shows how hey s ack up in
e e yday WBAN use. Fo each g oup, we include a able ha
spells ou how he algo i hms lea n, how complex hey a e,
how accu a e hey can be, and wha hey do well o poo ly.
We make i clea ha he ‘‘bes ’’ model depends on wha he
WBAN needs: can i scale, sa e ba e y powe , and keep up
wi h eal- ime demands? We also call ou he headaches ha
come wi h using ML in hese ne wo ks: hea y compu a ion,
lopsided da ase s, ha d- o-explain decisions, and e hical con-
ce ns. We inish by lagging he open ques ions and poin ing
o whe e u u e wo k could make ML an e en be e i
o heal hca e and o he WBAN applica ions. In sho , he
e iew o e s a s aigh o wa d oadmap o anyone looking
o see how machine lea ning is pushing WBAN echnology
o wa d. The 55 chosen s udies show ha an eno mous ange
o ML models a e being used and a e success ul in WBAN
applica ions. Figu e 13 is he axonomy o he ML-based
echniques used in WBANs ega ding his SLR.
1) ML-BASED ACTIVITY RECOGNITION TECHNIQUES FOR
WBANs
The s udy compa es di e en ML me hods o human ac i -
i y ecogni ion (HAR) and ela ed applica ions in heal hca e
WBANs. Key indings and de elopmen s in he ield a e high-
ligh ed as he analysis examines hei lea ning pa adigms,
model complexi y, pe o mance, and limi a ions. Because
hey can handle complex da a and ex ac high-dimensional
cha ac e is ics, DL echniques like CNNs and LSTM ne -
wo ks a e widely used. On he o he hand, supe ised
lea ning, which p o ides in e media e complexi y bu ypi-
cally alls sho o he accu acy a ained by DL models, is he
ounda ion o con en ional echniques like SVMs.
DL models con inuously pe o m well in e ms o accu-
acy and model complexi y. Fo ins ance, El-Adawi e
al, and Boga e al. [133] and [139] bo h use CNNs o
a ain ema kable accu acy, highligh ing he po ency o
hese a chi ec u es in ea u e ex ac ion. In a simila ein,
194748 VOLUME 13, 2025
A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
FIGURE 11. PRISMA low diag am.
FIGURE 12. Publica ion yea s o selec ed s udies.
Kedja e al. [135] combine CNN and LSTM o achie e
p ominen posi ion iden i ica ion and classi ica ion accu acy
in bo h line-o -sigh (LoS) and non-line-o -sigh (NLoS) si -
ua ions, bu in con as o DL models [119]. The SVM-based
VOLUME 13, 2025 194749

A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
FIGURE 13. Taxonomy o he ML-based echniques o WBANs.
me hod o e s in e media e complexi y and lowe accu acy,
bu i s ill shows gains in me ics like loss, delay, and h ough-
pu .
The models ha e signi ican limi a ions despi e hei
excellen accu acy. El-Adawi e al. [139] poin ou ha
in e p e abili y and senso -speci ic di icul ies a e no con-
side ed, al hough hey a e essen ial conside a ions in del-
ica e heal hca e sec o s. The pe o mance in he s udy by
Kedja e al. [135] is hampe ed by da a imbalance, espe-
cially ega ding NLoS loca ion p edic ion. T ea men delays
esul om olde me hods like SVMs’ incapaci y o accu-
a ely and p omp ly iden i y pa ien p oblems, as no ed by
Ka hu ia e al. in [119]. In simila esea ch by Boga e al.
in [133] highligh s di icul ies in modelling ac i e lea ning
pa adigms and e alua ing decision-making when wea able
senso s a e included.
No wi hs anding hese d awbacks, he models make no e-
wo hy con ibu ions. The model by El-adawi e al. in [139]
demons a e i s e ec i eness by achie ing accu acy, F-
measu e, and Ma hews Co ela ion Coe icien (MCC). The
s udy by Kedja e al. in [135] demons a e i s eliabili y in
loca ion iden i ica ion asks by exhibi ing good classi ica-
ion accu acy in LoS-NLoS scena ios and a dec eased oo
mean squa e e o (RMSE) in channel p edic ion. Despi e i s
educed accu acy, Ka hu ia e al. [119] shows mino bene i s
in loss, delay, and h oughpu measu es. Las ly, he s udy by
Boga e al. in [133] emphasizes how well ea u e selec ion
wo ks o enhance HAR ecogni ion pe o mance. Table 6
depic s de ails o ML-based ac i i y ecogni ion echniques
o WBANs.
2) ML-BASED ANOMALY DETECTION TECHNIQUES FOR
WBANs
A compa ison o ML app oaches used o goals in heal h-
ca e WBANs is p esen ed in his sec ion. I emphasizes he
ML echniques, hei in icacy, lea ning models, p ecision,
cons ain s, and salien ea u es. The emphasis is on ackling
issues like ha m ul da a pa e ns in WBAN sys ems and
anomaly de ec ion in physiological pa ame e s.
The goal o Boga e al. [133] was c ea ing a echnique
o iden i ying abno mali ies in physiological da a cap u ed
by WBAN senso s. They used supe ised lea ning wi h
an ANN. The model’s du abili y in anomaly de ec ion is
demons a ed by i s medium complexi y and good accu acy,
p ecision, ecall, and F1 sco e. Ne e heless, he s udy poin s
ou d awbacks in assessing decision-making in HAR asks,
especially when combining wea able senso s wi h ac i e
lea ning pa adigms. None heless, he echnique wo ks well
o p o ide dependable pe o mance o de ec ing anomalies
in WBANs.
Finding dange ous da a pa e ns b ough on by senso mal-
unc ions, inaccu a e eadings, and possible hos ile ac i i y
was he main goal o [129]. They employed a DL s a egy
ha deli e s high complexi y and pe o mance by combining
194750 VOLUME 13, 2025
A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
TABLE 6. ML-Based ac i i y ecogni ion echniques o WBANs.
CNN and LSTM ne wo ks. The model’s e ec i eness in
managing a ious da a is demons a ed by i s high accu acy
measu e ac oss a ious subjec s in he da ase s. Howe e ,
he s udy highligh s new inconsis encies in WBANs and
di icul ies wi h huge da a s eams. Mo e complex solu ions
ha can handle dynamic and expansi e WBAN en i onmen s
a e equi ed, as cu en s a is ical and ML echniques e-
quen ly all sho in iden i ying such pa e ns. Table 7shows
he de ails o ML-based anomaly de ec ion echniques o
WBANs.
3) ML-BASED COMMUNICATION TECHNIQUES FOR WBANs
This sec ion p o ides an academic compa ison o ML
app oaches used in WBANs, along wi h in o ma ion on hei
ad an ages and disad an ages, and impo an pe o mance
indica o s such as h oughpu , ene gy e iciency, la ency,
eliabili y, and packe delay a io (PDR). Fe nandes e al.
[124] educed da a packe loss by 10% by using an ANN in
supe ised lea ning o inc ease communica ion eliabili y in
WBANs. Howe e , his me hod led o highe la ency because
o he dependabili y o he bes -a ained ac ion. CNN was
used by Liu e al. in [126] o DL-based pe o mance gains
in ene gy e iciency and eliabili y; howe e , he esul ’s
gene alizabili y is limi ed by he absence o comp ehensi e
eal-wo ld es ing. Chen e al. [140] used Q-lea ning in end-
ing o educe ene gy use; howe e , he e alua ion was based
only on simula ions, which migh no accu a ely e lec eal-
wo ld ci cums ances.
Al hough he ene gy needed o un DRL models may o se
some o he ad an ages in p ac ice, Gup a e al. [141] ound
ha DRL models imp o e ene gy e iciency in eal- ime
applica ions. While maximum ene gy consump ion was s ill
a p oblem, Ahmad e al. [127] disco e ed ha using CNN in
ou ing p o ocols imp o ed packe ansmission and decision-
making, dec easing pa h loss a ios and inc easing ene gy
e iciency. High-pe o mance anscei e s o WBANs we e
designed by Ali e al. in [142] using DL; ne e heless,
hei s a egy migh be cons ained by da a a es o up o
1.312 Mbps, which would no be sui able o high- h oughpu
applica ions.
Al hough o he c ucial aspec s like la ency and eliabili y
we e no ho oughly in es iga ed, He e al. [143] de eloped
a deep Q-lea ning-based powe con olle in RL ha ou -
pe o med baseline con olle s in e ms o ene gy e iciency.
While a s udy by Roy e al. in [144] emphasized ML’s
e olu iona y po en ial in au oma ing heal hca e p ocedu es,
hei indings o e looked c ucial ac o s, including e hical
conside a ions and legal cons ain s ha could hampe ML
adop ion in heal hca e.
Simila ly, He e al. in [143] poin ed ou ha al hough hei
me hod e ec i ely inc eased ne wo k li e ime and educed
packe loss a es, scaling up ne wo ks may be signi ican ly
hampe ed by he compu ing cos and ime equi ed o RL.
Al hough o he impo an aspec s like la ency and eliabil-
i y we e no ho oughly examined, RL by He e al. in [143]
p oposed a deep Q-lea ning-based powe con olle ha ou -
pe o med baseline con olle s in e ms o ene gy e iciency.
I highligh ed he e olu iona y po en ial o ML in au oma -
ing heal hca e p ocedu es bu ailed o conside key ac o s
such as mo al dilemmas and egula o y es ic ions ha could
p e en ML om being widely adop ed in he medical ield.
Roy e al. [144] emphasized ML’s ans o ma i e ole in
au oma ing heal hca e p ocesses. Howe e , hei indings
o e looked c i ical issues like e hical conce ns and legal
egula ions, which migh hinde ML adop ion in heal hca e.
Simila ly, a s udy by He e al. in [143] no ed ha while hei
model success ully ex ended ne wo k longe i y and educed
packe loss a es, he compu a ional cos and ime equi ed
o RL could pose a signi ican ba ie when scaling up
ne wo ks.
VOLUME 13, 2025 194751
A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
TABLE 7. ML-based anomaly de ec ion echniques o WBANs.
DRL was used o in elligen e lec ing su ace (IRS) aided
me hods by Xiao e al. [145] who poin ed ou ha e en while
he e a e ad an ages in e ms o ene gy e iciency and lowe
ea esd opping a es, he compu ing demands o he lea ning
p ocess may esul in signi ican ene gy o e head. Al hough
impo an aspec s like secu i y and scalabili y we e no ully
add essed, Rao e al. [146] demons a ed how RL could
dynamically modi y channel alloca ion in WBANs, boos -
ing dependabili y in medical da a ans e . The QL-based
app oach by Mohammadi e al. in [147] inc eased ne wo k
connec ion, ene gy e iciency, and has d awbacks such as
sleep/wake p oblems and eme gency packe loss. In ano he
esea ch by Kim e al. in [148] DRL enhanced WBAN pe -
o mance, emphasizing inc easing h oughpu while lowe ing
ene gy usage. They did poin ou ha i is s ill unclea how
well he sugges ed DRL s a egy will wo k wi h la ge -scale
WBAN ins alla ions. The complexi y o scheduling and
powe con ol may ise wi h he numbe o de ices in he
ne wo k, which could impac he sys em’s o e all pe o -
mance. To e alua e he sugges ed app oaches’ obus ness and
p ac ical applicabili y, he au ho s unde lined he necessi y o
addi ional alida ion h ough es ing in a ious en i onmen s
and eal-wo ld implemen a ions.
To ensu e QoS in heal hca e sys ems, Abdu e al. [12] used
ANN h ough supe ised lea ning o c ea e ene gy-conscious
ou ing o WBANs. Thei s udy did no e one d awback,
hough: he equen equi emen o eplace senso s because
o esou ce cons ain s. This could pose a undamen al p ob-
lem o sus aining sys em pe o mance o e he long un.
Despi e his, hei me hod showed signi ican ene gy manage-
men and ou ing e iciency gains, making i a iable op ion
o In e ne o Things-based medical applica ions.
DNNs we e used by Cwalina e al. in [128] o inc ease
communica ion e iciency by add essing he p oblems o LoS
and NLoS ca ego iza ion. None heless, when wo king wi h
lea ning da ase s, he ins abili y o he SVM and Th esh-
old Me hod (THM) limi ed he in es iga ion, which would
ha e a ec ed he classi ica ion’s esilience and dependabil-
i y. Despi e hese obs acles, hei DL me hod success ully
inc eased he classi ica ion e ec i eness in WBANs, indica -
ing i s po en ial o boos sys em pe o mance.
Meh ani e al. [131] educed da a communica ion and
inc eased ene gy e iciency in WBANs by combining uzzy
logic wi h LSTM ne wo ks. They acknowledged a d awback
in no in es iga ing o he po en ially e ec i e ML ech-
niques ha may ha e u he op imized he pe o mance,
e en hough hei me hod p oduced a s unning educ ion o
communica ed da a by 81% and ene gy expendi u e by 73%.
E en hough hei s a egy e ec i ely lowe ed ene gy usage,
including mo e algo i hms o handle o he aspec s o WBAN
op imiza ion would be ad an ageous.
Rega ding FL, Consul e al. [149] no ed ha while
ede a ed RL enhanced la ency, ene gy consump ion, and
h oughpu in WBANs, i may also esul in addi ional com-
pu ing o e head. Al hough hey o e looked he ene gy cos
o edge se e s handling o loaded asks, i is also wo h
no ing ha FL may inc ease communica ion and ene gy
consump ion because he senso nodes need o pe iodically
exchange model upda es o g adien s wi h a cen al se e .
Howe e , mos o he e iewed s udies did no p o ide key
in o ma ion, such as he size o he sha ed model, he deg ee
o comp ession used, o how o en upda es a e ansmi ed,
making i di icul o es ima e his o e head o de e mine i s
impac on ba e y li e unde IEEE 802.15.6 powe cons ain s.
Fu u e wo k should epo hese pa ame e s mo e clea ly so
ha he pe o mance and p ac icali y o FL-based WBANs
can be e alua ed mo e accu a ely.
Yin e al. [150] demons a ed how DQN in RL may op i-
mize he QoS in WBANs. Table 8p o ides an ML-based
communica ion app oach o WBANs.
4) ML-BASED ROUTING PROTOCOL TECHNIQUES FOR
WBANs
Wi h an emphasis on hei use cases, QoS enhancemen s,
d awbacks, and signi ican con ibu ions, his pa s udies ML
app oaches used in ou ing p o ocols o heal hca e WBANs.
194752 VOLUME 13, 2025
A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
TABLE 8. ML-based communica ion echniques o WBANs.
VOLUME 13, 2025 194753
A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
In p ac ice, hese me ics, h oughpu , la ency, ene gy use,
e o a es, and mo e, guide bo h esea che s and enginee s.
They e eal each design’s s eng hs and weak spo s and
se e as ya ds icks o imp o ing u u e WBAN ha dwa e
and p o ocols. A well- ounded se o benchma ks, sp ead
ac oss mul iple sub- hemes (scalabili y, eliabili y, ene gy
e iciency, da a quali y, and secu i y), le s us judge whe he
a sys em is uly eady o clinical use. By iewing WBAN
pe o mance h ough his mul i-lens app oach, we can keep
pushing he echnology o wa d and deli e mo e dependable,
pa ien - iendly heal h-moni o ing solu ions.
1) PERFORMANCE EVALUATION OF THE MODEL
This pa assesses he pe o mance o models used in
WBANs, including p edic ed accu acy capabili ies. E al-
ua ing his model’s pe o mance is c i ical o making
us wo hy and e ec i e decisions in heal hca e applica ions.
Howe e , when e alua ing he algo i hm’s o e all pe -
o mance in iden i ying DDoS a acks in cloud-assis ed
WBANs, he examined me ics o classi ica ion accu acy, ee
size, ime, and memo y a e essen ial [122].
In addi ion, he s udy, which in eg a es p io i y-based
adap i e scheduling wi h DRL o powe con ol, demon-
s a es supe io pe o mance in e ms o PDR and h ough-
pu [159]. Compa ed o adi ional me hods, i also dec eases
la ency and lowe s powe usage. Ano he esea ch by
Li e al. in [160] claims ha he DRL op imiza ion (DRL-
OPT) me hod e icien ly adjus s o en i onmen al changes,
con inuously achie ing excellen pe o mance wi h minimal
compu a ional cos .
a: QUANTITATIVE PERFORMANCE COMPARISONS OF THE
MODEL ON WBAN
In his SLR, we in es iga e he use o sophis ica ed ML
echniques in WBANs, ocusing on impo an pe o mance
measu es such as accu acy, la ency, and powe consump ion.
These ac o s a e essen ial o de e mining how well ML
models wo k in eal-li e heal h moni o ing si ua ions ha
need quick esponses and sma use o ene gy.
DL models such as CNNs and LSTM ne wo ks a e no able
o hei high accu acy, wi h a 97% success a e in p edic -
ing use beha iou . No only a e hese models accu a e, bu
hey also minimize da a ans e by 81%, hence lowe ing
la ency. Fu he mo e, hey inc ease ene gy e iciency by 15%
as compa ed o p e ious scheduling app oaches u ilized in
WBANs [126],[131],[132].
SVMs, on he o he hand, p oduce eliable indings a a
92.1% accu acy le el. While hey main ain minimal la ency
( ypically less han 20 milliseconds), hey a e less capable o
handling mo e complex inpu han DL models. None heless,
SVMs a e e y powe -e icien , making hem an excel-
len choice o ba e y-powe ed de ices. They consume less
ene gy han deep lea ning models, which usually demand
g ea e p ocessing powe [161].
RF and DT also pe o m well, pa icula ly wi h
high-dimensional senso da a, wi h an accu acy o 88.6%.
These models s ike a ai mix be ween accu acy and he
necessi y o eal- ime pe o mance, wi h low la ency (a ound
20 milliseconds). They also use 30% less powe han deep
lea ning models, making hem app op ia e o wea able
de ices ha mus conse e ene gy [120],[162].
Table 14 summa izes he pe o mance compa isons o a -
ious ML echniques used in WBANs, ocusing on accu acy,
la ency, and powe consump ion.
b: ENERGY-EFFICIENCY, RELIABILITY, AND LATENCY
Due o powe limi s and eal- ime da a ansmission needs,
WBANs equi e ene gy-e icien , eliable, and low-la ency
wea able de ices. Me ics in his sub- heme e alua e he
ene gy consump ion, eliabili y, and la ency o da a ansmis-
sion p o ocols and algo i hms o op imize esou ce u iliza ion
and imp o e sys em pe o mance. Ne e heless,
Table 15 indica es he de ails o his sub- heme.
c: REDUCTION IN ENERGY CONSUMPTION AND LATENCY
This sec ion ocuses on lowe ing la ency and ene gy con-
sump ion in WBANs. WBANs can inc ease esponsi eness
and ex end de ice ba e y li e by educing ene gy con-
sump ion and communica ion delays, ul ima ely imp o ing
use expe ience and sys em e iciency. The ou comes o he
sub- hemes a e shown in Table 16.
d: COMPUTATIONAL TIME
The ime needed o da a p ocessing, analysis, and
decision-making in WBANs is measu ed by compu a ional
ime. Since quick esponses a e essen ial o pa ien mon-
i o ing and in e en ion, eal- ime applica ions equi e a
minimum p ocessing ime. The sugges ed unsupe ised col-
o ing algo i hm (UCA) demons a ed con e gence ou comes
wi h bo h co ec and inaccu a e posi ioning in he s udy
by Ma e al. [137]. I pe o med be e han cu en ech-
niques in e ms o obus ness agains opological changes,
as e algo i hm con e gence, educed in e e ence s eng h,
and less ime complexi y. Likewise, in ano he s udy by
Li e al. [160] he DRL op imiza ion (DRL-OPT) algo i hm
e ec i ely ollows en i onmen al a ia ions and main ains
good pe o mance wi h low compu a ional complexi y.
e: PRECISION, ACCURACY, SENSITIVITY, AND SPECIFICITY
This combines 4 ele an sub- hemes: p ecision, o e all accu-
acy, sensi i i y, and speci ici y. The esul is shown in
Table 17. These me ics e alua e he p ecision, accu acy, sen-
si i i y, and speci ici y o diagnos ic and p edic i e models in
WBANs. Reliable illness iden i ica ion, ea men planning,
and heal hca e decision-making all depend on ele a ed le els
o p ecision and accu acy.
: THROUGHPUT
The a e o success ul da a ansmission in WBANs is quan i-
ied by h oughpu measu emen s, which show how well he
ne wo k can handle da a a ic. The h oughpu o a WBAN
194760 VOLUME 13, 2025

A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
TABLE 14. Summa izing he pe o mance compa isons o a ious ML echniques used in WBANs, ocusing on accu acy, la ency, and powe consump ion.
TABLE 15. Summa y o he ene gy-e icien , eliable, and low-la ency
sub- heme.
TABLE 16. Summa y o Reduc ion in Ene gy Consump ion and La ency
Sub- heme.
TABLE 17. Summa y o p ecision, accu acy, sensi i i y, and speci ici y
sub- heme.
ne wo k is deno ed as TH [165] as ollows:
TH =RchennelTda a
Tsimula ion
(1)
whe e Tchennal Is he licensed channel’s da a ansmission
a e Tda a Is he o al ansmission ime o all success ul
ansmissions, and he Tsimula ion Is he sys em simula ion
ime?
Th oughpu op imiza ion is c i ical o ensu ing ha senso
da a is deli e ed o heal hca e decision-suppo sys ems on
ime. Table 18 compa es he h oughpu be ween some mod-
els and he adi ional me hod.
TABLE 18. Compa ison in e ms o h oughpu .
g: MODEL PERFORMANCE IMPROVEMENT
This sec ion e alua ed he pe o mance imp o emen s made
by model op imiza ion and e inemen s a egies.Resea che s
ha e inc eased he p edic ion models’ e icacy and accu acy
in WBANs by i e a i ely imp o ing model pe o mance.
Thus, he au ho [127] p oposed an imp o ed quali y o ou -
ing p o ocol (IM-QRP) ha shows no able imp o emen s,
including a 10% inc ease in esidual ene gy, a 30% educ ion
in pa h loss a io, a 10% imp o emen in packe ansmission
eliabili y, and a 7% imp o emen in SNR, in con as o he
VOLUME 13, 2025 194761
A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
cu en coope a i e link ene gy-e icien balanced algo i hm
(CO-LEEBA) and QoS pa h ou ing disco e y (QPRD) ou -
ing p o ocols. Wi h 98.8% accu acy, 99.9% ecall, 97.9%
F-sco e, and 99.8% p ecision, he p oposed hyb id a i icial
NN and g asshoppe op imiza ion algo i hm (ANN-GOA)
o anomaly iden i ica ion ou pe o ms o he con en ional
echniques in a dis inc s udy [118].
Fu he mo e, he expe imen al indings showed a ema k-
ably low alse ala m a e and an accu acy o o e 96% o
p oblem de ec ion. The o e all e icacy o he echnique is
demons a ed by he aul managemen p o o ype, which uses
he sugges ed amewo k o classi y aul s, au oma e sensing
node p o iling, and make i easie o ain and alida e new
models [120].
h: HYPERPARAMETER TUNING
Measu es o hype pa ame e uning assess how well model
hype pa ame e s can be op imized o imp o ed pe o -
mance. Wi hin WBAN applica ions, adjus ing model pa ame-
e s can lead o no able imp o emen s in an icipa ed accu acy
and esilience.
Howe e , a high p edic ion a e o ca diac diso de s
is achie ed by he RNN lexible a chi ec u e and pa am-
e e uning, which a e made possible by he unica e
swa m-sail ish op imiza ion (TS-SFO). The sugges ed
model success ully aises p edic ion accu acy in he WBAN
se ing [125].
i: MODEL OUTPERFORMANCE
Measu es o model ou pe o mance o iden i y he mos e ec-
i e me hods o ce ain asks o applica ions. This sub- heme
e alua es he pe o mance o mul iple models o algo i hms
wi hin WBANs. Resea che s can ind he mos e ec i e
app oaches and solu ions o WBAN adop ion by con as ing
model pe o mance wi h indus y no ms o i al app oaches.
Ne e heless, Table 19 summa izes he esul s unde he
sub- heme model ou pe o mance.
j: SUMMARY OF THE SECTION
WBANs a e ans o ming hanks o ML, which has imp o ed
hei unc ionali y and add essed signi ican issues. F om
communica ion wi hin WBANs o in e ac ions be ween
WBANs and beyond, his pa examined he in eg a ion o
ML echniques ac oss se e al WBAN laye s. I has been
demons a ed ha ad anced ML models, including DL, RL,
Gen AI, and FL, enhance eal- ime p ocessing, ene gy e i-
ciency, da a quali y, and secu i y in WBANs. Robus anomaly
de ec ion, indi idualized heal hca e moni o ing, and e ec i e
esou ce managemen a e made possible by hese me hods.
Howe e , p i acy p o ec ion, scalabili y, and model in e -
p e abili y emain c ucial opics o u he s udy. This sec ion
demons a es how ML can ans o m WBANs and open
he doo o scalable, secu e, and pa ien -cen ed heal hca e
sys ems by analysing exis ing esea ch and poin ing ou un e-
sol ed issues. In addi ion o add essing echnical cons ain s,
ML in eg a ion in o WBANs mee s he g owing need o
in elligen , lexible heal hca e solu ions.
VI. DISCUSSION AND STATISTICS
The s udy ound ha supe ised lea ning was he mos used
ML model ac oss he analysed s udies, wi h 41% o he o al
me hods used. Supe ised ML echniques, such as, ANN,
SVM, DT, SVR, KNN, LRE, RF show how c ucial labelled
da a and supe ised lea ning algo i hms a e in de eloping
p edic ion models o classi ica ion and eg ession asks
in WBANs like e ec i e in classi ying aul s, au oma ing
sensing node p o iling, de ec ing hidden pa e ns in heal h-
ela ed da a, and con e gence ime, and accu acy o disease
p edic ion due o hei excep ional pe o mance in a ious
applica ions [111],[118],[119],[120],[121],[122],[156],
[167]. DL echniques a e pa o his lea ning echnique,
no able o hei abili y o au oma ically lea n and imp o e
da a ep esen a ions [111],[118],[119],[120],[121],[122],
[156],[167].
Con olu ional ne wo ks (CNN, DCNN, R-CNN) and
ecu en a chi ec u es (RNN, LSTM, Con LSTM, and Bi-
LSTM) a e wo examples o DL models ha ha e signi -
ican ly impac ed WBAN applica ions. DL has been shown
o imp o e ene gy e iciency, ex end ne wo k li e ime, lowe
ansmission powe , and lessen in e e ence a he sys em
le el [113],[114],[115]. I also imp o es communica ion
eliabili y by allowing o aul classi ica ion, au oma ic node
p o iling, and e icien pic u e comp ession and enc yp ion
models [117],[118],[119],[120],[121]. By educing alse
posi i es in disease de ec ion, unco e ing hidden pa e ns in
physiological da a, and achie ing high p edic ion a es in
applica ions like hea disease classi ica ion and ea ly-s age
lung cance diagnosis models, a hy hmia de ec ion, s ess-
le el p edic ion, and all de ec ion, DL suppo s accu a e
heal h moni o ing a he clinical le el [12],[122],[123],
[124],[125],[126],[127]. These esul s highligh he dual
ole o DL in enhancing heal hca e ou comes and ne wo k
pe o mance in WBAN sys ems. DL models as a subse
o supe ised ca ego y make he p opo ion o supe ised
lea ning app oaches become highe , con i ming ha supe -
ised echniques emain dominan in WBAN- ela ed ML
applica ions.
RL, which accoun ed o 35% o he da a, was he sec-
ond mos common ML model behind supe ised lea ning.
RL echniques ha e demons a ed encou aging ou comes in
op imizing esou ce alloca ion, ou ing p o ocols, and adap-
i e con ol mechanisms in WBANs [168],[169],[170],
[171]. These me hods enable agen s o expe imen wi h hei
su oundings and disco e he bes decision-making policies.
This echnique includes DRL, which is a hyb id echnique
ha uses DL and RL concep s. This combina ion o me hods
is pe ec o applica ions like adap i e esou ce alloca ion
and indi idualized heal hca e moni o ing in WBANs because
i can esol e complex op imiza ion p oblems and adjus o
shi ing en i onmen al a iables [168],[172]. In addi ion o
inc easing o e all use ulness o WBAN secu i y, imp o ing
194762 VOLUME 13, 2025
A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
TABLE 19. Summa y o he esul s o he sub- heme model ou pe o mance.
WBANs’ e ec i eness and dependabili y o medical appli-
ca ions, and eco ding a a ie y o ne wo k in e ac ions [141],
[145],[148],[155],[164],[173],[174],[175],[176].
The s udy demons a ed ha , in addi ion o hese DRL
me hods, specialized algo i hms such as he FL and FRL
a e also a pa o an RL amewo k ha is used o p o ide
collabo a i e ML ha educes ime delay, enhances h ough-
pu , lowe s ene gy consump ion, and p o ec s p i acy when
combining ML and WBAN da a [149].
Unsupe ised ML app oaches, which ha e shown p omise
in anomaly de ec ion and clus e ing analysis in WBANs,
we e applied in 24% o he eco ds. Wi hou he need o
labelled ins ances, hese me hods make i possible o ind
unde lying pa e ns and s uc u es in da a, and hey inc ease
he ne wo k li espan, enhance adap abili y o changing ne -
wo k opologies, educe in e e ence s eng h, and accele a e
algo i hm con e gence [137],[138],[152].
Figu e 14 depic s he de ails o he iden i ied ML in
WBANs om 2017–2025.
Figu e 15 shows he issues in WBANs ha ML add essed
be ween 2017 and 2025. The mos impo an p oblem ha
ML add esses in WBANs is communica ion, which ecei es
a signi ican amoun o a en ion (40%) and is c ucial o
dependable da a ans e and sys em pe o mance. Pe son-
alized heal hca e ecei es signi ican a en ion (25%) as
e idence o he expanding end owa d pa ien -cen e ed solu-
ions. Ano he c ucial a ea is secu i y (11%), which empha-
sizes how c ucial i is o p o ec p i a e heal h in o ma ion
and p ese e sys em in eg i y. Focusing on mode a e a eas,
like Ac i i y Recogni ion (8%) and QoS (8%), enhances he
e iciency and adap abili y o he sys em.
In con as , mo e specialized a eas such as Rou ing P o o-
cols (4%) and Anomaly De ec ion (4%) ecei e less a en ion
in WBANs esea ch, despi e hei signi icance. Body mobil-
i y, limi ed ene gy esou ces, and he need o ul a-low
la ency cons ain ou ing in WBANs, making he di ec appli-
ca ion o ML echniques challenging. Simila ly, anomaly
de ec ion equi es la ge, di e se, and labelled da ase s o
e ec i ely iden i y abno mal physiological pa e ns; how-
e e , hese da ase s a e o en di icul o ob ain due o p i acy
FIGURE 14. Dis ibu ion o machine lea ning app oaches applied in
WBAN esea ch (2017–2025), showing ha Supe ised Lea ning (41%)
domina es, ollowed by Rein o cemen Lea ning (35%) and Unsupe ised
Lea ning (24%).
conce ns and limi ed access o pa ien da a. Despi e hei
eal-wo ld impo ance, hese challenges explain he educed
ocus o machine lea ning in hese ields. We p opose one-
class/sel -supe ised and ede a ed me hods, enhanced wi h
p i acy-p ese ing da a syn hesis and unce ain y es ima-
ion, o anomaly de ec ion, as well as budge ed/sa e o
model-based RL and ligh weigh , g aph-awa e policies o
ou ing o add ess hese issues. To make esul s clinically
ele an , we also ecommend e alua ing c oss-subjec gene -
aliza ion and epo ing deploymen - ocused me ics (ene gy
pe decision, la ency, and alse ala ms).
This dis ibu ion demons a es a s a egy ha balances
a en ion o o he complemen a y a eas while gi ing
p io i y o communica ion, pe sonalized heal hca e, and
secu i y.
The way we moni o heal h may change i ML echniques
a e inco po a ed in o WBANs. Tiny wea able senso s, which
eed eal- ime da a o machine lea ning algo i hms, enable
physicians o iden i y ea ly symp oms o disease, pe sonalize
ea men plans, and moni o daily ac i i ies. I is di icul
o ansla e his ision in o a wo king sys em, acco ding
VOLUME 13, 2025 194763
A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
FIGURE 15. Issues in WBANs add essed by ML (2017–2025).
Communica ion (40%) is he la ges ocus, ollowed by Pe sonalized
Heal hca e (25%) and Secu i y (11%). Ac i i y Recogni ion (8%) and
Quali y o Se ice (QoS, 8%) ecei e mode a e a en ion, while Rou ing
P o ocols (4%) and Anomaly De ec ion (4%) a e compa a i ely
unde explo ed. O e all, he li e a u e p io i izes eliable connec i i y and
pa ien -cen ic ou comes, wi h a meaning ul—bu smalle —emphasis on
secu i y and specialized asks.’’
o se e al s udies. Fo example, well-known p oblems wi h
WBAN deploymen s include senso un eliabili y and da a
noise b ough on by elec ode displacemen , ba e y d ain,
o en i onmen al in e e ence [161],[177]. Simila ly, ene gy
limi a ions con inue o be c ucial: he majo i y o WBAN
de ices un on iny ba e ies, and sophis ica ed machine lea n-
ing models use mo e powe , which educes hei long- e m
use ulness [178].
Clinicians a e equen ly hesi an o us complex model
p edic ions wi hou clea , anspa en explana ions, so he
lack o model in e p e abili y has eme ged as a majo
obs acle o clinical adop ion, su passing ha dwa e limi a-
ions [171]. Addi ionally, since pe o mance is epo ed
using a a ie y o me ics (accu acy, la ency, ene gy, sensi-
i i y, e c.), he absence o s anda dized e alua ion ame-
wo ks makes i mo e di icul o compa e s udies, as no ed
by [136] and [179].
Finally, he exis ing WBAN li e a u e does no ade-
qua ely add ess he issues o scalabili y and c oss-popula ion
gene aliza ion [180].
I will ake de elopmen s in explainable ML laye s, s an-
da dized es beds, ul a-e icien algo i hms, and da a p e-
p ocessing o o e come hese obs acles. In WBAN se ings,
obus secu i y and p i acy ea u es like FL and ligh weigh
enc yp ion a e also essen ial o sa egua ding p i a e heal h
in o ma ion [157],[158]. Fo he c ea ion o scalable, in e -
p e able, sa e, and ene gy-e icien ML models ha a e
adap ed o WBAN cons ain s, mul idisciplina y coope a ion
be ween enginee s, compu e scien is s, and heal hca e p o-
essionals is he e o e c ucial.
A. DEVELOPED THEMES
Th ee p ima y hemes eme ged om he hema ic analy-
sis: (1) he mos common machine lea ning models used in
WBAN esea ch, (2) pe o mance e alua ion me ics and he
insigh s hey p o ide, and (3) upcoming esea ch di ec ions
and challenges, along wi h hei sub hemes such as accu-
acy, e iciency, and compu a ional e ec i eness. These we e
among he key pe o mance me ics employed o e alua e
WBAN models in he i s heme, Pe o mance E alua-
ion. Model pe o mance assessmen , ene gy-e icien and
la ency- educing designs, educ ion in ene gy consump ion,
compu a ion ime, accu acy, sensi i i y, speci ici y, h ough-
pu , pe o mance imp o emen s, hype pa ame e uning, and
model ou pe o ming o he s, along wi h educing ansmis-
sion cos s, enhancing senso unc ionali y, inc easing he
use ulness o WBANs, and making he 55 selec ed s udies
demons a e ha a wide a ie y o ML models a e being
applied in WBANs a e e ec i e and e icien . O he c i ical
a eas included sys em li espan, powe e iciency, in e e ence
mi iga ion, equency u iliza ion, du y cycle op imiza ion,
a ic educ ion, and eliable da a deli e y. Ano he signi i-
can issue was secu i y, wi h sub- hemes ocused on me hods
o p e en jamming a acks, dec ease block e o a es, and
ensu e QoS. Addi ionally, he e was p og ess in de eloping
ML-based mHeal h sys ems, imp o ing image enc yp ion
and comp ession, educing he isk o misdiagnosis and alse
posi i es, enabling adap i e pe o mance in shi ing WBAN
opologies, and enhancing p edic ion accu acy o hea
diseases.
By me hodically examining ML app oaches, ends, and
p e e ences, his heme p o ides us wi h in o ma ion abou
he mos e ec i e ways o apply ML in a ious WBAN con-
ex s. Addi ionally, i cla i ies he dis inc ion be ween simple
and sophis ica ed p ocesses. These h ee domains o e a
ho ough summa y o he cu en s a e o WBANs, highligh -
ing esea ch challenges, pe o mance equi emen s, and he
g oundb eaking po en ial o ML-based solu ions.
1) THEME 1: MACHINE LEARNING MODELS USED
The i s heme examined how ML algo i hms a e ans o m-
ing he cu en s a e o WBAN sys ems. The impo ance
o iden i ying equen ly used machine lea ning algo i hms
and unde s anding hei unc ions in handling he complex
and dynamic da a gene a ed by WBANs was unde lined
by his heme. A a ie y o app oaches ha e been s udied,
including mo e complex DL models as pa o supe ised
lea ning s a egies [111],[118],[119],[120],[121],[122],
o he mo e sophis ica ed DL models [12],[63],[120],
[121],[123],[124],[125],[126],[127],[128],[129],[130],
[131],[132],[133],[134],[135],[136] and con en ional
unsupe ised echniques [137],[138]. The ad an ages o
each kind o algo i hm inc ease he p ecision, e ec i eness,
and adap abili y o WBAN applica ions. Because hey a e
simple o use and comp ehend, basic ML echniques like
DT, SVMs, and KNN a e well-liked in esou ce-cons ained
194764 VOLUME 13, 2025
A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
en i onmen s. Con e sely, sophis ica ed ML me hods like
CNN, RNN, LSTM ne wo ks, and RL excel a handling
high-dimensional da a, iden i ying in ica e ea u es, and
s eamlining dynamic p ocesses like slo alloca ion and
ene gy managemen . The chosen s udies show ha an eno -
mous ange o ML models a e being used and a e success ul
in WBANs. These esul s show how impo an ML is o
os e ing c ea i i y and enhancing WBAN pe o mance.
In addi ion, his heme emphasizes how supe ised lea n-
ing is mos u ilized o maximize ene gy e iciency and
imp o e WBAN communica ion eliabili y, accoun ing o
41% o he e iewed s udies. Al hough less common, RL has
po en ial applica ions in dynamic esou ce alloca ion and
adap i e ou ing. A small bu inc easing pe cen age o s udies
use eme ging echniques like FL, especially in heal hca e
se ings whe e p i acy is a conce n. Acco ding o his dis i-
bu ion, he e is g owing expe imen a ion wi h decen alized
and adap i e models, e en hough adi ional supe ised
me hods con inue o se e as he ounda ion.
Fu he mo e, Table 20 p o ides an o e iew o he ML
models ound in WBAN esea ch be ween 2017 and 2025,
emphasizing he mos common me hodologies (e.g., supe -
ised, unsupe ised, and RL) and hei main uses, including
anomaly de ec ion, communica ion op imiza ion, and pe -
sonalized heal hca e. In addi ion o he use-case-speci ic
discussions p e iously p o ided, his o e s a comp ehensi e
summa y o he modelling echniques used.
2) THEME 2: PERFORMANCE EVALUATION
Accu acy, e iciency, and compu a ional e ec i eness we e
among he key pe o mance me ics used o e alua e WBAN
models in Theme 2 ‘‘Pe o mance E alua ion’’. Model pe -
o mance e alua ion, ene gy-e icien and la ency- educing
design, ene gy consump ion educ ion, compu a ion ime,
accu acy, sensi i i y, speci ici y, h oughpu , pe o mance
imp o emen , hype pa ame e adjus men , and model ou pe -
o mance we e some o he sub- hemes. The s anda ds by
which WBAN sys ems and ML-based models a e e alua ed,
enhanced, and alida ed a e comp ised o hese sub- hemes.
Howe e , he e alua ion o pe o mance in WBAN-ML
s udies is co e ed unde his heme. The mos used me ic is
ene gy e iciency, emphasizing he impo ance o ba e y li e
o wea able echnology. While heal hca e applica ions ocus
on accu acy, sensi i i y, and speci ici y, communica ion-
ela ed s udies p io i ize h oughpu and la ency. Addi ion-
ally, some esea ch examines model ou pe o mance and
hype pa ame e uning as key e alua ion aspec s. This case
s udy highligh s he di e si y o pe o mance issues in he
ield by showing ha e alua ion me ics a e no uni o mly
applied bu depend on he speci ic WBAN applica ion.
3) THEME 3: CHALLENGES AND EMERGING PATHWAYS
This heme examined clinical and echnical issues ha a e
impeding he ad ancemen o WBAN applica ions. This ca e-
go y consis s o 21 sub- hemes, including educing ansmis-
sion cos s, imp o ing he unc ionali y o senso s, inc easing
he use ulness o WBANs, and making sys ems mo e depend-
able and e icien . The sys em’s li espan, powe e iciency,
in e e ence educ ion, equency u iliza ion, du y cycle op i-
miza ion, a ic educ ion, and eliable da a deli e y we e
o he c i ical a eas. Ano he majo issue was secu i y, wi h
sub- hemes concen a ing on me hods o p e en jamming
a acks, educe block e o a es, and ensu e QoS. The de el-
opmen o ML-based mHeal h sys ems, imp o ed image
enc yp ion and comp ession, educed isk o misdiagnosis
and alse posi i es, adap i e pe o mance in shi ing WBAN
opologies, and imp o ed p edic ion accu acy o hea dis-
eases we e also included in his heme.
Hence, he heme sugges s ha applying ML echniques
o WBANs p esen s a complex en i onmen wi h nume -
ous oppo uni ies and challenges. Iden i ying and esol ing
hese issues and u u e esea ch di ec ions a e c ucial o
maximizing he e ec i eness o ML echniques in WBAN
applica ions. Wi hin his discou se, we delinea e pi o al
challenges and p opose p ospec i e ajec o ies o u u e
esea ch.
VII. OPEN CHALLENGES AND FUTURE RESEARCH
DIRECTIONS FROM THE INTEGRATION OF ML IN WBANs
Al hough p omising, inco po a ing ML in o WBANs o e s
a dis inc collec ion o di icul ies and opens new esea ch
di ec ions. These di icul ies include handling he a ied and
dynamic na u e o WBAN en i onmen s, coping wi h he
di e si y in human physiology, gua an eeing da a p i acy and
secu i y, and p ocessing da a in eal- ime wi h cons ained
compu e esou ces [111]. ML models s ill s uggle o pe -
o m eliably ac oss di e en pa ien g oups and clinical
scena ios. A good i s s ep is simply acknowledging hese
gaps and se ing an agenda o deepe s udy. P omising
di ec ions include s onge da a-enc yp ion me hods, ul a-
ligh weigh lea ning algo i hms, and adap i e models ha
imp o e each ime new da a a i es.
Poin ing ou hese sho alls and laying ou a oadmap o
u u e wo k is c i ical i we wan WBAN-based ML sys ems
o deli e on hei p omise o be e heal h moni o ing and
decision-making [79]. Resea che s mus ackle se e al p ac-
ical hu dles: sca ce aining da a, he need o me ge eco ds
om many sou ces, and he day- o-day challenges o chang-
ing es ablished clinical ou ines [185],[186]. Add essing
hese issues should aise he e iciency, accu acy, and usabil-
i y o ML-d i en moni o ing ools [181]. I we o e come
hem, ML-enabled WBANs could ans o m heal hca e by
suppo ing con inuous, pe sonalized ea men ha imp o es
pa ien ou comes [187].
In eg a ing ML in o WBANs also p oduces a as -
mo ing, complex landscape. Sol ing oday’s p oblems and
iden i ying omo ow’s esea ch ques ions will make hese
echniques a mo e use ul in p ac ice. One pe sis en issue is
noisy, highly a iable senso da a. Human mo emen , de ice
aul s, and changes in he su ounding en i onmen can all
VOLUME 13, 2025 194765

A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
TABLE 20. Summa y o Machine lea ning models in WBANs Resea ch (2017-2025). Highligh s he p edominan machine lea ning app oaches and hei
p ima y applica ions, including communica ion op imiza ion, anomaly de ec ion, and pe sonalized heal hca e.
dis o eadings, educing model accu acy. Two ad anced
app oaches, au oencode s o noise educ ion and RL o
adap i e sensing, show eal p omise. Au oencode s clean he
aw s eam and pull ou he mos in o ma i e ea u es, while
RL adjus s sampling pa ame e s on he ly o sui cu en
condi ions [177]. Robus anomaly-de ec ion pipelines and
good da a p e-p ocessing a e s ill essen ial o eed he model
eliable inpu .
Finally, since mos wea ables un on small ba e ies, ene gy
e iciency is a cons an conce n. Demanding ML wo kloads
194766 VOLUME 13, 2025
A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
can quickly d ain powe and sho en se ice li e. Fu u e
esea ch mus conside powe budge s, econciling model
complexi y wi h he p ac ical cons ain s o wea able de ices.
Ligh weigh ML models like MobileNe and model com-
p ession s a egies like p uning ha e been c ea ed o lessen
his. These me hods p ese e model pe o mance while
lowe ing he compu a ional o e head [178]. In addi ion,
FL minimizes he ene gy cen alized p ocessing uses by
enabling dispe sed compu a ion ac oss se e al de ices.
By u ilizing hese s a egies, WBANs can ex end ba e y li e
wi hou sac i icing unc ionali y.
Because WBANs handle highly sensi i e heal h in o -
ma ion, keeping ha da a secu e and p i a e is essen ial.
A b each du ing ansmission o s o age could ha e se i-
ous consequences o pa ien s. FL helps educe his isk by
keeping aw da a on each pe son’s de ice; only he upda es
om he ained model a e sha ed, so he o iginal in o ma ion
ne e lea es he pa ien ’s con ol. Fu he mo e, GANs and
GenAI can gene a e p i acy-p ese ing da ase s o aining
ML models [157],[158].
As shown in Table 13, mos exis ing ML-based secu i y
app oaches in WBANs concen a e on coun e ing spoo ing,
ampe ing, and in o ma ion leakage a he communica ion
laye . In con as , e y li le a en ion has been gi en o
sa egua ding he lea ning models hemsel es. None o he
e iewed s udies di ec ly add esses he isk o model poison-
ing o Byzan ine beha io in ede a ed o dis ibu ed WBAN
se ings. This gap sugges s a need o u u e wo k on obus
agg ega ion s a egies, us -awa e ede a ed lea ning, and
secu e alida ion o model upda es, o main ain he in eg i y
and eliabili y o ML models used in con inuous heal hca e
moni o ing applica ions.
Blockchain-based au hen ica ion sys ems, decen alized
da a p ocessing, and homomo phic enc yp ion a e all iable
op ions o secu e compu a ions. These solu ions enhance
he secu i y and p i acy o WBAN sys ems while boos ing
s akeholde and use con idence.
Clea and unde s andable ML models o WBANs a e
necessa y o heal hca e p ac i ione s o us and u ilize
hese solu ions in clinical. The equen lack o ans-
pa ency in black-box models’ explana ions o o ecas s may
make hem ha de o accep . To add ess his issue, XAI
me hodologies such as Local In e p e able Model-agnos ic
Explana ions (LIME) and SHapley Addi i e exPlana ions
(SHAP) ha e been o mula ed. By showing how models
make decisions, hese me hods make hem clea e [188].
Fu he mo e, a en ion-based models make i easie o unde -
s and by showing which pa s o he decision-making p ocess
a e mos impo an . These ad ancemen s a e c ucial o he
in eg a ion o ML in o medical p ac ices.
Fo p omp heal hca e ac ions, eal- ime p ocessing is
c ucial, especially in impo an si ua ions like iden i ying
i egula i ies o an icipa ing c ises. Howe e , in dynamic con-
ex s, ypical ML models could ha e ouble wi h la ency
and adap a ion. Low-la ency DNNs o RL in conjunc ion
wi h edge compu ing ha e been p oposed o emedy his
issue. Th ough da a p ocessing locally on he de ice, edge
compu ing elimina es he delays ha come wi h sending da a
o dis an se e s [189]. These me hods allow o eal- ime
analysis and lexibili y, gua an eeing ha WBANs can eac
quickly o changing si ua ions.
When used in WBANs, ML models mus unc ion eli-
ably ac oss pa ien demog aphics and medical si ua ions.
Poo gene aliza ion may esul om a ia ions in da a dis-
ibu ion and en i onmen al ac o s. T ans e lea ning and
me a-lea ning ha e been p esen ed as solu ions o his p ob-
lem. Me a-lea ning eaches a model o how o lea n, so i
can pick up new ac i i ies o se ings wi h e y li le ex a
aining. T ans e lea ning le s a model ha was ained in one
job adap quickly o a di e en one [180]. Teaming up se e al
models in an ensemble makes p edic ions s u die and mo e
accu a e, an ad an age when you scale up iny bio-nano sen-
so s o wi eless b ain anscei e s, whe e eliabili y ma e s
mos .
P og ess is slowe han i should be, hough, because
esea che s s ill don’ ha e a common ya ds ick o judging
ML ools in WBANs. We need a sha ed amewo k ha
weighs accu acy, esponse ime, ba e y use, and how easy
he model is o in e p e , all a once. A mul i-me ic sco eca d
like ha would spo ligh ade-o s, s ee design choices, and
make i possible o compa e esul s ac oss s udies [179]. Wi h
s anda dized benchma ks in place, he ield could inno a e
and imp o e much as e .
Howe e , he e a e ad an ages and disad an ages o in e-
g a ing ML in o WBANs. To ully exploi he po en ial o
ML-d i en WBANs, speci ic solu ions a e needed o da a
quali y, ene gy e iciency, secu i y, in e p e abili y, eal- ime
p ocessing, scalabili y, and e alua ion me ics. Ad anced
ML echniques such as XAI, FL au oencode s, and edge
compu ing p o ide p omising app oaches o ackling hese
issues. By implemen ing hese concep s, esea che s and
p ac i ione s may imp o e WBANs’ usabili y, e iciency, and
dependabili y, imp o ing pa ien ca e. Table 21 lis s he main
ML op ions now used o ackle co e WBAN challenges.
Beyond algo i hmic imp o emen s, he in eg a ion o ML
in WBANs will hea ily ely on inno a ions in sensing and
wi eless communica ion ha dwa e o suppo mo e accu a e
and adap i e physiological moni o ing.
A. EMERGING MICROWAVE SENSING AND
COMMUNICATION TECHNOLOGIES FOR INTELLIGENT
WBANs
The inco po a ion o ML in WBANs will p og essi ely ely
on enhancemen s in sensing and communica ion ha dwa e
capable o deli e ing mo e comp ehensi e and depend-
able da a. Fu u e WBAN sys ems a e an icipa ed o u ilize
mic owa e-based senso s ha unc ion on he p inciples o
dielec ic pe u ba ion, whe ein a ia ions in biological is-
sue pe mi i i y esul in quan i iable al e a ions in esonan
equency and quali y ac o . These senso s, which use dual-
equency, meande ed mic os ip, and subs a e-in eg a ed
VOLUME 13, 2025 194767
A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
TABLE 21. Summa y o he ML op ions o add ess a ious p oblems in WBANs.
wa eguide a chi ec u es, a e be e o con inuous physio-
logical moni o ing because hey a e mo e accu a e, smalle ,
and mo e lexible [8],[9],[190],[191],[192],[193].
These ad ancemen s c ea e no el esea ch oppo uni ies o
machine lea ning models ha can handle high- equency
biomedical da a, execu e ea u e ex ac ion, and adjus o
eal- ime physiological luc ua ions.
Meanwhile, in WBAN en i onmen s, nex -gene a ion
wi eless communica ion echnologies, like subs a e-
in eg a ed and leaky-wa e an enna sys ems, will enable
low-la ency, in e e ence- esis an , and ene gy-e icien da a
ansmission [10],[11],[194],[195]. Fu u e s udies should
concen a e on c oss-laye in eg a ion, in which machine
lea ning algo i hms dynamically engage wi h he commu-
nica ion and sensing laye s o accomplish in elligen ene gy
managemen , adap i e slo alloca ion, and op imal da a low.
The e o e, new de elopmen s in wi eless and mic owa e
echnology will be c ucial o he de elopmen o in elligen ,
sel -op imizing WBAN a chi ec u es d i en by ML.
B. TOWARD STANDARDIZED BENCHMARKS FOR ML
EVALUATION IN WBANs
A key challenge ound ac oss he e iewed s udies is he
absence o s anda dized da ase s and e alua ion p o ocols
o es ing ML models in WBANs. Mos s udies depend on
p i a e o inconsis en da a ga he ed unde a ying expe i-
men al condi ions, which makes i ha d o ep oduce esul s
o compa e models accu a ely. This inconsis ency a ec s pe -
o mance epo ing o key pa ame e s, including accu acy,
la ency, and ene gy consump ion.
Recen s udies in wea able compu ing and HAR ha e
shown ha open and uni ied benchma king amewo ks
imp o e esea ch compa abili y and anspa ency [196],
[197],[198]. Publicly a ailable da ase s such as Mobile
Heal h (MHEALTH), Wea able S ess and A ec De ec ion
(WESAD), Physical Ac i i y Moni o ing (PAMAP2), and
Massachuse s Ins i u e o Technology–Be h Is ael Hospi al
(MIT-BIH) A hy hmia al eady p o ide di e se physiological
and mo ion da a ha a e ele an o WBAN applica ions.
Howe e , hese da ase s a e o en used inconsis en ly ac oss
s udies.
To o e come his limi a ion, u u e wo k should ocus on
de eloping an open-sou ce WBAN-ML benchma k ame-
wo k ha includes:
1) Da ase in eg a ion: A cu a ed collec ion o ep e-
sen a i e public WBAN da ase s (e.g., MHEALTH,
WESAD, PAMAP2, MIT-BIH).
2) S anda d p ep ocessing pipeline: Common da a clean-
ing, segmen a ion, and no maliza ion s eps o ensu e
consis en inpu ac oss models.
3) Uni ied e alua ion me ics: A ixed se o indica o s
such as F1-sco e, la ency (ms), ene gy pe in e -
ence (mJ), model size (kB), and in e p e abili y sco e
(XAI-sco e) o enable objec i e pe o mance
compa ison.
4) Communi y leade boa d: A sha ed pla o m o
esea che s o upload esul s and compa e models
unde he same expe imen al se ings.
C ea ing such a s anda dized benchma k would enhance
ep oducibili y, ai ness, and collabo a ion in WBAN-ML
esea ch. I would also accele a e he de elopmen o eliable,
ene gy-e icien , and explainable ML models o eal- ime
heal hca e moni o ing.
194768 VOLUME 13, 2025
A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
FIGURE 16. Explainabili y–Accu acy Pa e o on ie o ML models in eg a ed in o WBANs o heal hca e applica ions. The plo isualizes he ade-o
be ween model explainabili y ( anspa ency) and p edic i e accu acy ac oss di e en ML pa adigms. La ge bubble sizes indica e a highe equency o
occu ence in he e iewed s udies (2017–2025). Models such as DL and DRL achie e high accu acy bu low explainabili y, whe eas uzzy logic (FL),
unsupe ised, and supe ised ML app oaches p o ide be e in e p e abili y wi h mode a e accu acy.
C. EXPLAINABILITY–ACCURACY TRADE-OFF IN ML
MODELS FOR WBAN HEALTHCARE
A key challenge in applying machine lea ning wi hin
WBAN-based heal hca e sys ems is inding an
app op ia e balance be ween p edic i e accu acy and in e -
p e abili y. Deep lea ning models—including CNNs, RNNs,
and newe hyb id deep ein o cemen lea ning app oaches—
o en achie e ex emely high pe o mance, wi h epo ed
F1-sco es equen ly abo e 95%. Howe e , hese models
ope a e as ‘‘black boxes,’’ making i di icul o clinicians
o unde s and o e i y how decisions a e being made. On he
o he hand, mo e anspa en models such as decision ees,
andom o es s, and uzzy logic sys ems allow he easoning
p ocess o be examined and alida ed, hough hey may o e
sligh ly lowe p edic i e accu acy.
Figu e 16 illus a es his ade-o by showing he
accu acy–explainabili y Pa e o on ie de i ed om he
55 s udies e iewed. The igu e indica es ha deep lea ning
and deep RL app oaches clus e in he egion o highes
accu acy bu lowes in e p e abili y, whe eas me hods like
decision ees, andom o es s, and uzzy logic occupy a
middle g ound whe e bo h in e p e abili y and pe o mance
a e balanced. This obse a ion is consis en wi h ecen
li e a u e s essing ha explainabili y is essen ial o clin-
ical us and accep ance in medical IoT and WBAN se -
ings [121],[139],[182].
Looking o wa d, esea ch should mo e owa d mul i-
objec i e op imiza ion s a egies ha explici ly accoun o
bo h accu acy and in e p e abili y. This may in ol e e al-
ua ing models using in e p e abili y me ics such as SHAP
o LIME alongside con en ional pe o mance measu es like
F1-sco e and p ecision. Es ablishing such dual e alua ion
s anda ds will suppo he de elopmen o machine lea n-
ing sys ems ha a e no only accu a e bu also anspa en
and eliable o eal- ime heal hca e moni o ing in WBAN
en i onmen s.
D. CLINICAL VALIDATION PATHWAYS, PATIENT
COMPLIANCE, AND REGULATORY CONSIDERATION
1) CLINICAL VALIDATION PATHWAYS
Clinical alida ion es s WBANs’ sa e y and pe o mance in
hei in ended en i onmen . The p ocess includes p eclinical
VOLUME 13, 2025 194769
A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
[155] G. Chen, X. Liu, M. Sho uzzaman, A. Ka ime, Y. Wang, and Y. Qi,
‘‘MEC-based jamming-aided an i-ea esd opping wi h deep ein o ce-
men lea ning o WBANs,’’ ACM T ans. In e ne Technol., ol. 22, no. 3,
pp. 1–17, Aug. 2022, doi: 10.1145/3453186.
[156] M. M. Islam and M. Shamshuzzoha, ‘‘Secu ing wi eless body a ea
ne wo ks da a ansmission wi h machine lea ning: A c oss- ie ame-
wo k o anomaly de ec ion and in usion p e en ion,’’ Compu .
S uc u al Bio echnol. Rep., ol. 2, Jun. 2025, A . no. 100031, doi:
10.1016/j.csb .2025.100031.
[157] G. Ace o, F. Giampaolo, C. Guida, S. Izzo, A. Pescapè, F. Piccialli, and
E. P ezioso, ‘‘Syn he ic and p i acy-p ese ing a ic ace gene a ion
using gene a i e AI models o aining ne wo k in usion de ec ion sys-
ems,’’ J. Ne w. Compu . Appl., ol. 229, Sep. 2024, A . no. 103926, doi:
10.1016/j.jnca.2024.103926.
[158] C. Uddagi i and B. V. Isunu i, ‘‘E hical and p i acy challenges o gene -
a i e AI,’’ in S udies in Compu a ional In elligence. Singapo e: Sp inge ,
2024, pp. 219–244, doi: 10.1007/978-981-97-8460-8_11.
[159] J. Kim, B. Kim, C. You, and H. Pa k, ‘‘P io i y based adap i e slo ed
ALOHA me hod using Q-lea ning,’’ J. Ko ean Ins . Commun. In . Sci.,
ol. 48, no. 3, pp. 350–358, Ma . 2023, doi: 10.7840/kics.2023.48.3.350.
[160] S. Li, H.-C. Yang, F. Xu, H. Hu, and F. Hu, ‘‘Ene gy-e icien elay ans-
mission o WBAN: Ene gy consump ion minimizing design wi h hyb id
supe ised/ ein o cemen lea ning,’’ IEEE In e ne Things J., ol. 11,
no. 10, pp. 17770–17779, May 2024, doi: 10.1109/JIOT.2024.3361772.
[161] A. P. Kunal and B. T. Saga , ‘‘C i ical heal hca e assessmen using
WBAN and SVM,’’ In . J. Inno . Technol. Explo ing Eng., ol. 10, no. 9,
pp. 84–86, Jul. 2021, doi: 10.35940/iji ee.i9362.0710921.
[162] N. Bilandi, H. K. Ve ma, and R. Dhi , ‘‘An in elligen and ene gy-e icien
wi eless body a ea ne wo k o con ol co ona i us ou b eak,’’ A abian J.
Sci. Eng., ol. 46, no. 9, pp. 8203–8222, Sep. 2021, doi: 10.1007/s13369-
021-05411-2.
[163] R. Mohammadi and Z. Shi mohammadi, ‘‘DRDC: Deep ein o ce-
men lea ning based du y cycle o ene gy ha es ing body sen-
so node,’’ Ene gy Rep., ol. 9, pp. 1707–1719, Dec. 2023, doi:
10.1016/j.egy .2022.12.138.
[164] Y. Chen, S. Han, G. Chen, J. Yin, K. N. Wang, and J. Cao, ‘‘A deep
ein o cemen lea ning-based wi eless body a ea ne wo k o loading op i-
miza ion s a egy o heal hca e se ices,’’ Heal h In . Sci. Sys ., ol. 11,
no. 1, p. 8, Jan. 2023, doi: 10.1007/s13755-023-00212-3.
[165] Z. Sad eddini, Ö. Makul, T. Ça da , and F. B. Günay, ‘‘Pe o mance anal-
ysis o licensed sha ed access based seconda y use s ac i i y on cogni i e
adio ne wo ks,’’ in P oc. Elec ic Elec on., Compu . Sci., Biomed. Eng.
Mee ing (EBBT), Ap . 2018, pp. 1–4, doi: 10.1109/EBBT.2018.8391442.
[166] N. Sha ma, H. Chadha, K. Singh, B. M. Singh, and N. Pa hak, ‘‘A no el
hyb id clus e ing based ansmission p o ocol o wi eless body a ea ne -
wo ks,’’ Compu ., Ma e . Con inua, ol. 69, no. 2, pp. 2459–2473, 2021,
doi: 10.32604/cmc.2021.014305.
[167] P. Khosh agh , J. Tan ee , A. M. Rahmani, M. Mohammadi, A. Meh an-
zadeh, J. Lansky, and M. Hosseinzadeh, ‘‘H-TERF: A hyb id app oach
combining uzzy mul i-c i e ia decision-making echniques and enhanced
andom o es o imp o e WBAN-IoT,’’ In e ne Things, ol. 32,
Jul. 2025, A . no. 101613, doi: 10.1016/j.io .2025.101613.
[168] D. P. Q. Ca nei o, A. A. Ca doso, and F. H. T. Viei a, ‘‘Adap i e esou ce
alloca ion in 5G CP-OFDM sys ems using Ma ko ian model-based ein-
o cemen lea ning algo i hm,’’ Neu al Compu . Appl., ol. 35, no. 13,
pp. 9421–9435, May 2023, doi: 10.1007/s00521-023-08406-2.
[169] S. Suknum, C. Thoasi i, and N. Jinapo n, ‘‘Q-lea ning-based
esou ce alloca ion in he e ogeneous cellula ne wo ks,’’ in P oc.
In . Elec . Eng. Cong . (iEECON), Ma . 2022, pp. 1–3, doi:
10.1109/iEECON53204.2022.9741639.
[170] V. A una, L. Anjaneyulu, and C. Bha , ‘‘Deep-Q ein o cemen lea ning
based esou ce alloca ion in wi eless communica ion ne wo ks,’’ in P oc.
IEEE In . Symp. Sma Elec on. Sys . (iSES), Dec. 2022, pp. 66–72, doi:
10.1109/ISES54909.2022.00025.
[171] Z. Zheng, S. Jiang, R. Feng, L. Ge, and C. Gu, ‘‘Su ey o ein o cemen -
lea ning-based MAC p o ocols o wi eless ad hoc ne wo ks wi h a
MAC e e ence model,’’ En opy, ol. 25, no. 1, p. 101, Jan. 2023, doi:
10.3390/e25010101.
[172] Y.-H. Li, Y.-L. Li, M.-Y. Wei, and G.-Y. Li, ‘‘Inno a ion and challenges
o a i icial in elligence echnology in pe sonalized heal hca e,’’ Sci. Rep.,
ol. 14, no. 1, p. 18994, Aug. 2024, doi: 10.1038/s41598-024-70073-7.
[173] B.-S. Kim, K.-I. Kim, and B. Shah, ‘‘BANSIM: A new disc e e-e en sim-
ula o o wi eless body a ea ne wo ks wi h deep ein o cemen lea ning
in Py hon,’’ J. Sys . A chi ., ol. 126, May 2022, A . no. 102489, doi:
10.1016/j.sysa c.2022.102489.
[174] L. Wang, Z.-H. You, D.-S. Huang, and F. Zhou, ‘‘Combining high
speed ELM lea ning wi h a deep con olu ional neu al ne wo k ea u e
encoding o p edic ing p o ein-RNA in e ac ions,’’ IEEE/ACM T ans.
Compu . Biol. Bioin ., ol. 17, no. 3, pp. 972–980, May 2020, doi:
10.1109/TCBB.2018.2874267.
[175] H. Su, M.-S. Pan, H. Chen, and X. Liu, ‘‘MDP-based MAC p o ocol o
WBANs in edge-enabled eHeal h sys ems,’’ Elec onics, ol. 12, no. 4,
p. 947, Feb. 2023, doi: 10.3390/elec onics12040947.
[176] B.-S. Kim, B. Shah, T. He, and K.-I. Kim, ‘‘A su ey on analy i-
cal models o dynamic esou ce managemen in wi eless body a ea
ne wo ks,’’ Ad Hoc Ne w., ol. 135, Oc . 2022, A . no. 102936, doi:
10.1016/j.adhoc.2022.102936.
[177] N. Raja hi, S. Di ya, S. A. Ela a asi, G. Sa i ha, and V. J. Ramya,
‘‘Adap i e in usion de ec ion in cybe -physical sys ems using ein-
o cemen lea ning-based au oencode s,’’ In . Con . In eg . In ell. Com-
mun. Sys . (ICIICS), ol. 2024, pp. 1–7, No . 2024, doi: 10.1109/ici-
ics63763.2024.10859561.
[178] P. V. Dan as, W. Sabino da Sil a, L. C. Co dei o, and C. B. Ca alho,
‘‘A comp ehensi e e iew o model comp ession echniques in machine
lea ning,’’ In . J. Speech Technol., ol. 54, no. 22, pp. 11804–11844,
No . 2024, doi: 10.1007/s10489-024-05747-w.
[179] I. El Badisy, N. G a eo, M. Khalis, and R. Gio gi, ‘‘Mul i-me ic compa -
ison o machine lea ning impu a ion me hods wi h applica ion o b eas
cance su i al,’’ BMC Med. Res. Me hodology, ol. 24, no. 1, p. 191,
Aug. 2024, doi: 10.1186/s12874-024-02305-3.
[180] A. Ve o uzzo, M.-R. Bouguelia, J. Vanscho en, T. Rögn aldsson,
and K. San osh, ‘‘Ad ances and challenges in me a-lea ning:
A echnical e iew,’’ IEEE T ans. Pa e n Anal. Mach. In ell.,
ol. 46, no. 7, pp. 4763–4779, Jul. 2024, doi: 10.1109/TPAMI.2024.
3357847.
[181] R. Roy, S. Mukhe jee, M. M. Ba al, A. K. Badhan, and M. Ra ind a,
‘‘Challenges encoun e ed in he implemen a ion o machine lea ning in
he heal hca e indus y,’’ in P oc. In . Con . Mach. Lea n. Big Da a Anal.,
2023, pp. 377–386.
[182] R. N. L. S. Kalpana, A. K. Pa o, and D. N. Rao, ‘‘Finding
he e iciency o Con Bi-LSTM o e an icipa ion o ad e sa ies in
WBANs,’’ Recen Pa en s Eng., ol. 19, no. 1, pp. 1–18, Jan. 2025, doi:
10.2174/0118722121255695231008171935.
[183] T. Thama aimanalan and S. Ramalingam, ‘‘Enhancing anomaly de ec ion
in WBANs using hyb id deep lea ning and op imiza ion algo i hms,’’
Neu al Compu . Appl., ol. 37, no. 15, pp. 9223–9243, May 2025, doi:
10.1007/s00521-025-11061-4.
[184] A. K. Pipal and R. Jagadeesh Kannan, ‘‘G aph a en ion laye -based
WideResNe o e icien ad e sa ial a ack de ec ion in wi eless body
a ea ne wo ks,’’ In . Secu . J., A Global Pe spec i e, ol. 2025, pp. 1–15,
Ap . 2025, doi: 10.1080/19393555.2025.2493102.
[185] J. C. Hong, N. C. W. Eclo , S. J. S ephens, Y. M. Mowe y, and M. Pal a,
‘‘Implemen a ion o machine lea ning in he clinic: Challenges and
lessons in p ospec i e deploymen om he sys em o high in ensi y
E aLua ion du ing adia ion he apy (SHIELD-RT) andomized con-
olled s udy,’’ BMC Bioin ., ol. 23, no. S12, p. 408, Sep. 2022, doi:
10.1186/s12859-022-04940-3.
[186] I. Hussain and A. U. Islam, ‘‘Resea ch di ec ion owa d IoT-based
machine lea ning-d i en heal h moni o ing sys ems: A su ey,’’ in P oc.
ICCVBIC, 2023, pp. 541–555, doi: 10.1007/978-981-19-9819-5_39.
[187] S. Vyas and S. Gup a, ‘‘WBAN-based emo e moni o ing sys em u ilising
machine lea ning o heal hca e se ices,’’ In . J. Sys . Sys . Eng., ol. 13,
no. 1, p. 100, 2023, doi: 10.1504/ijsse.2023.129054.
[188] D. E. Ma hew, D. U. Ebem, A. C. Ikegwu, P. E. Ukeoma, and
N. F. Dibiaezue, ‘‘Recen eme ging echniques in explainable a i icial
in elligence o enhance he in e p e able and unde s anding o AI models
o human,’’ Neu al P ocess. Le ., ol. 57, no. 1, p. 16, Feb. 2025, doi:
10.1007/s11063-025-11732-2.
[189] A. Amzil, M. Hanini, and A. Zaaloul, ‘‘Modeling and analysis o LoRa-
enabled ask o loading in edge compu ing o enhanced ba e y li e in
wea able de ices,’’ Clus e Compu ., ol. 28, no. 3, p. 201, Jun. 2025,
doi: 10.1007/s10586-024-04925-2.
[190] S. Kiani and P. Rezaei, ‘‘Mic owa e subs a e in eg a ed wa eguide
esona o senso o non-in asi e moni o ing o blood glucose concen a-
ion: Low cos and painless ool o diabe ics,’’ Measu emen , ol. 219,
Sep. 2023, A . no. 113232, doi: 10.1016/j.measu emen .2023.113232.
[191] S. Kiani, P. Rezaei, and M. Fakh , ‘‘Dual- equency mic owa e es-
onan senso o de ec nonin asi e glucose-le el changes h ough he
inge ip,’’ IEEE T ans. Ins um. Meas., ol. 70, pp. 1–8, 2021, doi:
10.1109/TIM.2021.3052011.
194776 VOLUME 13, 2025

A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
[192] S. Kiani, P. Rezaei, and M. Fakh , ‘‘Real- ime measu emen o liq-
uid pe mi i i y h ough label- ee meande ed mic owa e senso ,’’
IETE J. Res., ol. 70, no. 5, pp. 4606–4616, May 2024, doi:
10.1080/03772063.2023.2231875.
[193] S. Kiani, P. Rezaei, M. Ka ami, and R. A. Sadeghzadeh, ‘‘Band-s op
il e senso based on SIW ca i y o he non-in asi e measu ing o blood
glucose,’’ IET Wi eless Senso Sys ., ol. 9, no. 1, pp. 1–5, Feb. 2019, doi:
10.1049/ie -wss.2018.5044.
[194] P. Soh abi, P. Rezaei, S. Kiani, and M. Fakh , ‘‘A symme ical
SIW-based leaky-wa e an enna wi h con inuous beam scanning om
backwa d- o- o wa d h ough b oadside,’’ Wi eless Ne w., ol. 27, no. 8,
pp. 5417–5424, No . 2021, doi: 10.1007/s11276-021-02798-6.
[195] S. Kiani, P. Rezaei, and M. Fakh , ‘‘A CPW- ed wea able an enna a ISM
band o biomedical and WBAN applica ions,’’ Wi eless Ne w., ol. 27,
no. 1, pp. 735–745, Jan. 2021, doi: 10.1007/s11276-020-02490-1.
[196] M. Kase is, I. Kos a elis, and S. Malassio is, ‘‘A comp ehensi e su -
ey on deep lea ning me hods in human ac i i y ecogni ion,’’ Mach.
Lea n. Knowl. Ex ac ion, ol. 6, no. 2, pp. 842–876, Ap . 2024, doi:
10.3390/make6020040.
[197] R. Saman a, B. Saha, S. K. Ghosh, and R. Babu Roy, ‘‘Op imizing
TinyML: The impac o educed da a acquisi ion a es o ime se ies
classi ica ion on mic ocon olle s,’’ 2024, a Xi :2409.10942.
[198] S. Da idash illy, M. Ca dei, M. Hssayeni, C. Chi, and B. Gho aani,
‘‘Deep neu al ne wo ks o wea able senso -based ac i i y ecogni ion
in Pa kinson’s disease: In es iga ing gene alizabili y and model com-
plexi y,’’ Biomed. Eng. OnLine, ol. 23, no. 1, pp. 1–24, Feb. 2024, doi:
10.1186/s12938-024-01214-2.
[199] C. Venka esh, L. Si ayamini, M. V. Dasu, D. S. Vyshna i, D. Yasaswini,
D. S eeni asulu, G. Su yashanka a a ap asadad, and C. Naga aju, ‘‘Ca -
diac diagnosis sys em o hea diseases classi ica ion based on deep
lea ning and op imiza ion s a egies using ECG signals,’’ in P oc. In .
Con . Compu ., Elec . Commun. Eng. (ICCECE), Feb. 2025, pp. 1–8, doi:
10.1109/iccece61355.2025.10940447.
[200] K. Pe ez, M. Izha , A. Ahmed, N. Al u ki, and S. Abdullah, ‘‘Sma
implan able de ices o ca diac heal h: A no el sel -powe ed wi e-
less ECG moni o ing sys em using ene gy ha es ing and machine
lea ning-d i en anomaly de ec ion,’’ Sma Heal h, ol. 37, Sep. 2025,
A . no. 100582, doi: 10.1016/j.smhl.2025.100582.
[201] S. Ka hika and K. J. Gnanasel i, ‘‘A e iew o o ensics secu i y haza ds
and challenges in WBAN and heal hca e sys ems,’’ in Secu i y, P i acy,
and T us in WBANs and E-Heal hca e. Singapo e: Sp inge , No . 2024,
pp. 63–82.
[202] R. Abi ami and C. Mala hy, ‘‘Secu i y equi emen s and challenges in
WBANs and e-heal h sys ems,’’ in Secu i y, P i acy, and T us in WBANs
and E-Heal hca e, No . 2024, pp. 3–22.
[203] P ema ke No i ica ion 510(K) | FDA. Accessed: Jul. 10, 2025.
[Online]. A ailable: h ps://www. da.go /medical-de ices/p ema ke -
submissions-selec ing-and-p epa ing-co ec -submission/p ema ke -
no i ica ion-510k
ABDU IBRAHIM ADAMU ecei ed he Bache-
lo o Science deg ee (Hons) in compu e science
om Kano Uni e si y o Science and Technology
(cu en ly changed o Dango e Uni e si y o Sci-
ence and Technology), Kano, Nige ia, in 2014, and
he M.Sc. deg ee in compu e science wi h he Uni-
e si i Teknologi MARA, Shah Alam, Selango ,
Malaysia, in 2023, whe e he is cu en ly pu su-
ing he Ph.D. deg ee in elec ical enginee ing.
His esea ch in e es s include wi eless communi-
ca ion, a i icial in elligence, machine lea ning, big da a analysis, and cloud
compu ing.
PRAVEEN KUMAR DONTA (Senio Membe ,
IEEE) ecei ed he Bachelo o Technology and
Mas e o Technology deg ees (Hons.) om he
Depa men o Compu e Science and Enginee -
ing, JNTUA, Anan hapu , in 2012 and 2014,
espec i ely, and he Ph.D. deg ee om he Depa -
men o Compu e Science and Enginee ing,
Indian Ins i u e o Technology (Indian School o
Mines), Dhanbad, in June 2021. He is cu en ly
an Associa e P o esso (Docen ) wi h he Depa -
men o Compu e and Sys ems Sciences, S ockholm Uni e si y, Sweden.
F om July 2021 o June 2024, he was wi h he Dis ibu ed Sys ems G oup,
TU Wien, as a Pos doc o al Resea che . He was a Visi ing Ph.D. Fellow wi h
he Mobile and Cloud Labo a o y, Uni e si y o Ta u, Es onia, om July
2019 o Janua y 2020. His cu en esea ch is on lea ning-d i en dis ibu ed
compu ing con inuum sys ems, casual and conscious con inuum sys ems,
and in elligen da a p o ocols. He is an ACM P o essional Membe . He is an
Edi o ial Boa d Membe o IEEE INTERNET OF THINGS JOURNAL,Compu ing
(Sp inge ), ETT Wiley, POLS One,Measu emen , and Compu e Communi-
ca ions (Else ie ).
DARMAWATY MOHD ALI ecei ed he deg ee
(Hons.) in elec ical, elec onic, and sys ems engi-
nee ing om he Uni e si i Kebangsaan Malaysia
(UKM), in 1999, he mas e ’s deg ee om he Uni-
e si i Teknologi Malaysia (UTM), and he Ph.D.
deg ee om Uni e si i Malaya (UM), in 2012. She
is an Associa e P o esso wi h he Facul y o Elec-
ical Enginee ing, Uni e si i Teknologi MARA
(UiTM). She s a ed he i s job as a P oduc
Enginee . He esea ch in e es s include wi eless
access echnology and he p o ision o quali y o se ice (QoS) in wi eless
ne wo ks.
SOHAIL SARANG (Senio Membe , IEEE)
ecei ed he B.Eng. deg ee in elecommunica ion
enginee ing om Hamda d Uni e si y, Ka achi,
Pakis an, in 2014, he M.Sc. deg ee in elec ical
and elec onics enginee ing om he Uni e si i
Teknologi PETRONAS, Malaysia, in 2018, and
he Ph.D. deg ee in elec ical and compu e engi-
nee ing om he Facul y o Technical Sciences,
Uni e si y o No i Sad, Se bia. He is a Pos doc-
o al Resea che wi h he Depa men o Elec ical
Enginee ing, Facul y o Technical Sciences, Uni e si y o No i Sad.
His esea ch in e es s include ene gy ha es ing communica ions, low-
powe senso ne wo ks, ba e y- ee IoT, MAC p o ocols, and machine
lea ning-d i en communica ion algo i hms and p o ocols.
VOLUME 13, 2025 194777
A. I. Adamu e al.: Sys ema ic Li e a u e Re iew o Ad anced Machine Lea ning Techniques
GORAN M. STOJANOVIĆ (Membe , IEEE)
ecei ed he B.Sc., M.Sc., and Ph.D. deg ees in
elec ical enginee ing om he Facul y o Tech-
nical Sciences (FTS), Uni e si y o No i Sad
(UNS), Se bia, in 1996, 2003, and 2005, espec-
i ely. He is cu en ly a Full P o esso wi h
FTS, UNS. He has 27 yea s o expe ience in
esea ch and de elopmen . He has o e 18 yea s
o expe ience in w i ing, implemen ing, and coo -
dina ing EU- unded p ojec s (Ho izon Eu ope,
H2020, EUREKA, ERASMUS, and CEI), wi h a o al budge exceeding
22.86 MEUR. He supe ised 14 Ph.D. s uden s, 40 M.Sc. s uden s, and
60 diploma s uden s a FTS-UNS. He is he au ho /co-au ho o 280 a i-
cles, including 180 in pee - e iewed jou nals wi h impac ac o s, i e
books, h ee pa en s, and wo chap e s in a monog aph. He was a keyno e
speake a 14 in e na ional con e ences. His esea ch in e es s include
senso s, lexible elec onics, ex ile elec onics, edible elec onics, and
mic o luidics.
SUZI SEROJA SARNIN ecei ed he bache-
lo ’s deg ee in elec ical and elec onics om he
Uni e si i Teknologi Malaysia, in 1999, and he
M.Sc. deg ee in mic oelec onics and he Ph.D.
deg ee in elec ical enginee ing om Uni e si i
Kebangsaan Malaysia, in 2005 and 2018, espec-
i ely. F om 1999 o 2001, she was a Quali y
Con ol Enginee wi h Memo y Tech (M) Sdn.
Bhd. In Ma ch 2001, she was a Con ac Lec u e
wi h he Uni e si i Teknologi MARA. She con-
inued o wo k o he Depa men o Elec ical Enginee ing, Uni e si i
Teknologi MARA, as a Senio Lec u e . She has collabo a ed ac i ely wi h
esea che s in se e al o he elec ical enginee ing disciplines and indus ies.
He esea ch in e es s include wi eless communica ion, mul iple access,
space- ime coding and coding heo ies, signal p ocessing, and he In e ne
o Things.
194778 VOLUME 13, 2025