Co esponding au ho : Chekkala Tanisha.
Copy igh © 2025 Au ho (s) e ain he copy igh o his a icle. This a icle is published unde he e ms o he C ea i e Commons A ibu ion License 4.0.
Li e a u e su ey on machine lea ning app oaches o sleep diso de diagnosis
Ka i ha Soppa i , Chekkala Tanisha *, Anki Ve ma and Yedida Sai S ika
Depa men o Compu e Science Enginee ing (A i icial In elligence and Machine Lea ning), ACE Enginee ing College,
India.
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 2264-2270
Publica ion his o y: Recei ed on 03 Ap il 2025; e ised on 11 May 2025; accep ed on 13 May 2025
A icle DOI: h ps://doi.o g/10.30574/wja .2025.26.2.1856
Abs ac
Accu a e diagnosis o sleep diso de s, such as insomnia and sleep apnea, is c ucial o imp o ing heal h and well-being.
T adi ional diagnos ic me hods ely on expe analysis, which can be ime-consuming and p one o e o s. This P ojec
aims o op imize machine lea ning app oaches o enhance sleep diso de classi ica ion using he Sleep Heal h and
Li es yle Da ase . Va ious p ep ocessing echniques, including ea u e selec ion and da a balancing, will be used o
imp o e model pe o mance. Mul iple classi ie s will be e alua ed, wi h ensemble me hods such as G adien Boos ing
and Vo ing achie ing he highes accu acy. The P ojec aims o op imiza ion in machine lea ning echniques in
p edic ing sleep diso de s, o e ing a scalable and e icien solu ion o ea ly diagnosis and pe sonalized heal h
ecommenda ions.
Keywo ds: Sleep Diso de ; Machine Lea ning; Fea u e Selec ion; Ensemble Me hods; Ea ly Diagnosis
1. In oduc ion
Sleep diso de s, such as insomnia and sleep apnea, signi ican ly impac an indi idual's heal h, p oduc i i y, and o e all
well-being. Ea ly and accu a e diagnosis is essen ial o p e en long- e m heal h complica ions, ye adi ional
diagnos ic me hods o en ely on manual assessmen s ha can be ime-consuming, cos ly, and p one o human e o .
Wi h he g owing ad ancemen s in a i icial in elligence, machine lea ning has eme ged as a powe ul ool o
au oma ing and op imizing medical diagnoses. This p ojec aims o use op imized machine lea ning echniques o
imp o e he classi ica ion and p edic ion o sleep diso de s. By le e aging he Sleep Heal h and Li es yle Da ase ,
a ious da a p ep ocessing me hods, including ea u e selec ion and da a balancing, will be applied o enhance model
pe o mance. The p ojec will use mul iple machines lea ning classi ie s, ocusing on ensemble me hods such as
G adien Boos ing and Vo ing, which ha e shown p omise in achie ing high accu acy. The ul ima e goal o his p ojec
is o de elop an e icien , scalable, and eliable app oach o sleep diso de diagnosis. By e ining machine lea ning
models, his p ojec aims o con ibu e o ea ly de ec ion, enabling pe sonalized heal h ecommenda ions and
imp o ing o e all sleep heal h managemen .
2. Li e a u e e iew
2.1. Ti le- "Applying Machine Lea ning Algo i hms o he Classi ica ion o Sleep Diso de s"
Au ho : Alshamma i (2024) [1]
This s udy in es iga es he applica ion o a ious machine lea ning (ML) and deep lea ning models o classi y sleep
diso de s using he Sleep Heal h and Li es yle Da ase . The da ase comp ises 400 eco ds wi h 13 ea u es ela ed o
sleep pa e ns and daily ac i i ies. To enhance model pe o mance, Gene ic Algo i hms (GAs) we e employed o ea u e
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 2264-2270
2265
selec ion and hype pa ame e uning. GAs a e op imiza ion echniques inspi ed by na u al selec ion, use ul o
iden i ying op imal solu ions in la ge sea ch spaces. The s udy e alua ed se e al classi ie s: k-Nea es Neighbou s (k-
NN), Suppo Vec o Machines (SVMs), Decision T ees, Random Fo es s, and A i icial Neu al Ne wo ks (ANNs). Among
hese, ANNs achie ed he highes classi ica ion accu acy o 92.92%, ou pe o ming adi ional ML models. ANNs a e
compu a ional models inspi ed by biological neu al ne wo ks, capable o cap u ing complex nonlinea ela ionships in
da a. The esul s unde sco e he impo ance o ea u e selec ion and model op imiza ion in imp o ing diagnos ic
accu acy o sleep diso de s. The s udy concludes ha in eg a ing GAs wi h ML models can signi ican ly enhance he
classi ica ion o sleep diso de s, o e ing a p omising app oach o au oma ed diagnosis.
2.2. Ti le- "Reliable Au oma ic Sleep S age Classi ica ion Based on Hyb id In elligence"
Au ho : Shao e al. (2024) [2]
Shao and colleagues de eloped a hyb id in elligen model combining da a-d i en and knowledge-d i en app oaches o
au oma ic sleep s age classi ica ion. U ilizing he ISRUC and Sleep-EDFx da ase s, which include EEG and EOG
eco dings, he s udy employed a Tempo al Fully Con olu ional Ne wo k (TFCN) o cap u e empo al dependencies in
he da a. TFCNs a e deep lea ning a chi ec u es designed o sequence modeling, e ec i e in p ocessing ime-se ies
da a like EEG signals. Addi ionally, a mul i- ask ea u e mapping s uc u e was implemen ed o enhance he model's
abili y o lea n sha ed ep esen a ions ac oss asks. The model achie ed Mac o-F1 sco es o 0.804 on ISRUC and 0.780
on Sleep-EDFx, indica ing obus pe o mance ac oss mul iple sleep s ages. The in eg a ion o expe knowledge wi h
da a-d i en me hods allowed he model o co ec inconsis encies in sleep s age anno a ions, imp o ing eliabili y o
clinical applica ions. This hyb id app oach demons a es he po en ial o combining di e en o ms o in elligence o
enhance he accu acy and obus ness o sleep s age classi ica ion sys ems.
2.3. Ti le - "Classi ica ion o Sleep Diso de s Using Random Fo es on Sleep Heal h and Li es yle Da ase "
Au ho : Hidaya (2024) [3]
Hidaya 's s udy ocuses on classi ying sleep diso de s using he Random Fo es algo i hm applied o he Sleep Heal h
and Li es yle Da ase . Random Fo es is an ensemble lea ning me hod ha cons uc s mul iple decision ees du ing
aining and ou pu s he mode o hei p edic ions, enhancing accu acy and con olling o e i ing. The da ase was P e
p ocessed, and ea u es we e selec ed based on hei ele ance o sleep diso de s. The Gini Index was used o measu e
he quali y o spli s in he decision ees, helping o iden i y he mos in o ma i e ea u es. The model achie ed an
accu acy o 88%, demons a ing i s e ec i eness in classi ying sleep diso de s. Fu he analysis included examining
class dis ibu ions and ea u e co ela ions, p o iding insigh s in o ac o s in luencing sleep diso de s. The s udy
highligh s he obus ness o Random Fo es in handling complex da ase s and i s po en ial applica ion in heal hca e o
ea ly de ec ion and diagnosis o sleep- ela ed issues.
2.4. Ti le - "Sleep Diso de s De ec ion and Classi ica ion Using Random Fo es Algo i hm"
Au ho : Ta eq (2024) [4]
Ta eq's esea ch p esen s an au oma ed me hod o de ec ing and classi ying sleep diso de s, speci ically insomnia and
sleep apnea, using he Random Fo es algo i hm. The Sleep Heal h and Li es yle Da ase se ed as he da a sou ce,
con aining a ious ea u es ela ed o sleep pa e ns and heal h indica o s. The s udy in ol ed de eloping a machine
lea ning pipeline ha included da a p ep ocessing, ea u e selec ion, model aining, and e alua ion. Hype pa ame e
uning was conduc ed o op imize he model's pe o mance, ensu ing be e gene aliza ion o unseen da a. The Random
Fo es model achie ed an accu acy o 88%, ou pe o ming o he adi ional ML classi ie s. This app oach demons a es
he easibili y o eplacing adi ional expe -based diagnosis me hods wi h au oma ed models, po en ially educing
cos s and imp o ing e iciency in clinical se ings. The s udy unde sco es he impo ance o ensemble lea ning
echniques in de eloping eliable diagnos ic ools o sleep diso de s.
2.5. Ti le - "Iden i ica ion o C ucial Fac o s in Sleep Quali y Using Machine Lea ning Models and MRMR
Fea u e Selec ion"
Au ho : Wa unlawan e al. (2023) [5]
Wa unlawan and colleagues in es iga ed he impac o li es yle and medical ac o s on sleep quali y using machine
lea ning models combined wi h Minimum Redundancy Maximum Rele ance (MRMR) ea u e selec ion. MRMR is a
ea u e selec ion me hod ha aims o selec ea u es wi h maximum ele ance o he a ge a iable and minimal
edundancy among hemsel es. The Sleep Heal h and Li es yle Da ase was u ilized, and h ee key p edic o s we e
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 2264-2270
2266
iden i ied: physical ac i i y, sys olic blood p essu e, and BMI. A Bagged T ees classi ie , an ensemble me hod ha
combines mul iple decision ees ained on di e en subse s o he da a, was employed o classi ica ion. The model
achie ed an accu acy o 91.90%, indica ing high p edic i e pe o mance. The s udy emphasizes he signi ican ole o
li es yle changes in imp o ing sleep heal h and demons a es he e ec i eness o combining ea u e selec ion
echniques wi h ensemble lea ning models in heal hca e analy ics.
2.6. Ti le - "De ec ion and P edic ion o Sleep Diso de s by Co e Bed-In eg a ed RF Senso s"
Au ho : Zhang e al. (2023) [6]
Zhang e al. in oduced a no el app oach o de ec ing and p edic ing sleep diso de s using co e bed-in eg a ed adio
equency (RF) senso s. The s udy ocused on espi a o y dis u bances du ing sleep, such as sleep apnea, and u ilized
Nea -Field Cohe en Sensing (NCS) o cap u e con inuous espi a o y wa e o ms wi hou he use 's awa eness. Da a
we e collec ed om 27 pa ien s, and espi a o y ea u es we e ex ac ed o ain a Random Fo es machine lea ning
model. The model achie ed a sensi i i y o 88.6% and speci ici y o 89.0% in de ec ing apneic e en s, wi h he abili y o
p edic such e en s up o 90 seconds in ad ance. This non-in asi e me hod o e s a p omising al e na i e o adi ional
polysomnog aphy, p o iding eal- ime moni o ing and ea ly in e en ion capabili ies. The s udy highligh s he
po en ial o in eg a ing RF sensing echnology wi h machine lea ning o e ec i e and unob usi e sleep diso de
diagnos ics.
2.7. Ti le - "A Hie a chical App oach o he Diagnosis o Sleep Diso de s Using Con olu ional Recu en
Neu al Ne wo k"
Au ho : Wadicha e al. (2023) [7]
Wadicha and colleagues de eloped a hie a chical classi ica ion sys em o diagnosing sleep diso de s using
Con olu ional Recu en Neu al Ne wo ks (CRNNs). The s udy u ilized EEG da a om he CAP Sleep Da abase, ocusing
on analyzing Cyclic Al e na ing Pa e ns (CAP) phases. CRNNs combine con olu ional laye s, which ex ac spa ial
ea u es, wi h ecu en laye s, which cap u e empo al dependencies, making hem sui able o ime-se ies da a like
EEG signals. The model achie ed 91.45% accu acy in dis inguishing be ween heal hy and unheal hy CAP sequences and
90.55% accu acy in classi ying speci ic diso de s such as PLM, RBD, NFLE, NARCO, and INS. The esea ch demons a ed
ha ocusing on Phase B o CAP sequences signi ican ly imp o ed classi ica ion accu acy. This hie a chical app oach
showcases he e ec i eness o deep lea ning models in cap u ing complex pa e ns in physiological da a o accu a e
sleep diso de diagnosis.
2.8. Ti le- "P edic ion o Demen ia Based on Olde Adul s’ Sleep Dis u bances Using Machine Lea ning"
Au ho : Nyholm e al. (2023) [8]
Nyholm and colleagues explo ed he ela ionship be ween sleep dis u bances and demen ia isk in olde adul s using
machine lea ning echniques. The s udy analyzed da a om 4,175 pa icipan s aged 60 and abo e om he Swedish
Na ional S udy on Aging and Ca e in Blekinge (SNAC-B). Fi e machine lea ning algo i hms we e employed: G adien
Boos ing, Logis ic Reg ession, Gaussian Nai e Bayes, Random Fo es , and Suppo Vec o Machine (SVM). Each
algo i hm u ilized 10- old s a i ied c oss- alida ion, and pe o mance was e alua ed using me ics like B ie sco e and
ea u e impo ance. G adien Boos ing eme ged as he mos accu a e model, achie ing 92.9% accu acy, 0.926 F1-sco e,
and 0.974 ROC AUC. Signi ican p edic o s included day ime sleep du a ion exceeding wo hou s, sex, educa ion le el,
age, nigh ime awakenings, and sno ing. The s udy concluded ha sleep dis u bances a e associa ed wi h demen ia and
ha machine lea ning models can e ec i ely p edic demen ia isk, highligh ing he impo ance o ea ly in e en ion
s a egies.
2.9. Ti le - "Au oma ed Classi ica ion o Mul i-Class Sleep S ages Using a 9-Laye 1D-CNN"
Au ho : Sa apa hy and Logana han (2023) [9]
Sa apa hy and Logana han p oposed a deep lea ning model o mul i-class sleep s age classi ica ion using a 9-laye one-
dimensional Con olu ional Neu al Ne wo k (1D-CNN). The model was designed o p ocess polysomnog aphy (PSG)
signals, including EEG, ECG, and EOG da a. 1D-CNNs a e e ec i e in ex ac ing ea u es om ime-se ies da a, making
hem sui able o analyzing physiological signals. The a chi ec u e included mul iple con olu ional laye s ollowed by
pooling and ully connec ed laye s, enabling he model o lea n hie a chical ep esen a ions o he inpu signals. The
model ou pe o med adi ional me hods, achie ing highe accu acy in classi ying a ious sleep s ages. The s udy
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 2264-2270
2267
highligh ed he po en ial o deep lea ning app oaches in imp o ing he accu acy and e iciency o au oma ed sleep
diso de diagnosis, o e ing a scalable solu ion o clinical applica ions.
2.10. Ti le - "Ensemble SVM Me hod o Au oma ic Sleep S age Classi ica ion"
Au ho : Alicko ic and Subasi (2018) [10]
Roy and colleagues p oposed a hyb id machine lea ning model o de ec ing sleep apnea e en s using EEG signals. The
s udy u ilized da a om he PhysioNe Sleep-EDF Da abase, which con ains de ailed EEG eco dings anno a ed wi h
sleep apnea e en s. To p ep ocess he EEG signals, he au ho s applied band-pass il e ing and segmen a ion echniques
o isola e ele an equency bands such as del a, he a, alpha, and be a, which a e known o be associa ed wi h di e en
sleep s ages. Fea u e ex ac ion me hods like wa ele ans o m and s a is ical measu es (mean, a iance, en opy)
we e used o con e aw EEG signals in o a s uc u ed da ase sui able o ML analysis .The p oposed model combines
wo s ages: ea u e selec ion using P incipal Componen Analysis (PCA), ollowed by classi ica ion using a combina ion
o Suppo Vec o Machine (SVM) and k-Nea es Neighbo s (k-NN). SVM is a supe ised lea ning algo i hm e ec i e o
bina y classi ica ion asks, especially when dealing wi h high-dimensional da a, while k-NN is a simple, ins ance-based
lea ne ha classi ies a da a poin based on he majo i y o e o i s neighbo s. This ensemble app oach helped enhance
he model's gene aliza ion capabili ies .The hyb id model achie ed a classi ica ion accu acy o 94.6%, wi h high
sensi i i y and speci ici y, ou pe o ming adi ional models used in p e ious s udies. The esea ch demons a es ha
combining complemen a y algo i hms can yield obus and accu a e esul s in biomedical signal classi ica ion. Roy e
al. concluded ha hei model has p ac ical po en ial o in eg a ion in o eal- ime moni o ing sys ems o ea ly
de ec ion o sleep apnea, especially in non-clinical se ings.
Table 1 Compa ison able o he li e a u e
S .
No
Au ho Name
Ti le
Me hodology
Findings om Resea ch
Pape
1
Alshamma i
Applying Machine Lea ning
Algo i hms o he
Classi ica ion o Sleep
Diso de s
Used Sleep Heal h and Li es yle
Da ase , applied Gene ic
Algo i hms o op imiza ion,
es ed ANNs, Random Fo es ,
and SVM
ANNs achie ed he highes
accu acy o 92.92%,
su passing adi ional ML
models
2
Shao e al.
Reliable Au oma ic Sleep
S age Classi ica ion Based
on Hyb id In elligence
Used ISRUC and Sleep-
EDFx da ase s, applied TFCN
and mul i- ask ea u e mapping
Achie ed 0.804 Mac o-F1
(ISRUC) and 0.780 Mac o-
F1 (Sleep-EDFx), imp o ing
sleep s aging consis ency
3
Hidaya
Classi ica ion o Sleep
Diso de s Using Random
Fo es on Sleep Heal h and
Li es yle Da ase
Used Random Fo es wi h Gini
Index o ea u e anking
Achie ed 88% accu acy,
highligh ing he s eng h o
ensemble lea ning
echniques
4
Ta eq
Sleep Diso de s De ec ion
and Classi ica ion Using
Random Fo es Algo i hm
Used Random Fo es wi h
hype pa ame e uning on
Sleep Heal h and Li es yle
Da ase
Achie ed 88% accu acy,
sugges ing au oma ion can
eplace expe diagnosis
5
Wa unlawan
e al.
Iden i ica ion o C ucial
Fac o s in Sleep Quali y
Using Machine Lea ning
Models and MRMR Fea u e
Selec ion
Used MRMR ea u e selec ion
wi h Bagged T ees classi ie
Iden i ied key p edic o s
(physical ac i i y, sys olic
BP, BMI) and achie ed
91.90% accu acy
6
Zhang e al.
De ec ion and P edic ion o
Sleep Diso de s by Co e
BIRF senso s
Used RF senso s and Random
Fo es o analyze b ea hing
pa e ns
Achie ed 88.6% sensi i i y
and 89.0% speci ici y,
p edic ing apnea episodes
up o 90s in ad ance
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 2264-2270
2268
7
Wadicha e al.
A Hie a chical App oach o
he Diagnosis o Sleep
Diso de s Using
Con olu ional Recu en
Neu al Ne wo k
Used CRNNs wi h CAP Sleep
Da abase o hie a chical
classi ica ion
Achie ed 91.45% accu acy
in CAP sequence
classi ica ion and 90.55%
in diso de classi ica ion
8
Nyholm e al.
P edic ion o Demen ia
Based on Olde Adul s’
Sleep Dis u bances Using
Machine Lea ning
Used Random Fo es , Logis ic
Reg ession, G adien Boos ing,
SVM, and Nai e Bayes on SNAC-
B da ase
G adien Boos ing achie ed
he highes accu acy
(92.9%), linking sleep
dis u bances o demen ia
isk
9
Sa apa hy and
Logana han
Au oma ed Classi ica ion o
Mul i-Class Sleep S ages
Using a 9-Laye 1D-CNN
Used 9-laye 1D-CNN on PSG
signals (EEG, ECG, EOG)
Achie ed supe io
accu acy in mul i-class
sleep diso de
classi ica ion.
10
Alicko ic and
Subasi
Ensemble SVM Me hod o
Au oma ic Sleep S age
Classi ica ion
De eloped an ensemble SVM
model wi h EEG and ECG da a
Ensemble models imp o ed
sleep s age classi ica ion
accu acy o e single
classi ie s
The abo e able p esen s a compa a i e o e iew o en no able esea ch pape s add essing sleep diso de
classi ica ion using a ious machine lea ning (ML) and deep lea ning (DL) models. The s udies encompass a b oad ange
o algo i hms, da ase s, and objec i es, o e ing a holis ic iew o cu en ad ancemen s in he ield.
Among he s udies, he highes accu acy (92.92%) was achie ed by he pape i led "Applying Machine Lea ning
Algo i hms o he Classi ica ion o Sleep Diso de s" using ANN wi h Gene ic Algo i hm, signi ying he e ec i eness o
op imized ea u e selec ion and neu al ne wo ks in classi ying sleep diso de s. Simila ly, G adien Boos ing used in he
s udy on p edic ing demen ia om sleep pa e ns ("P edic ion o Demen ia Based on Olde Adul s’ Sleep Dis u bances")
also yielded an imp essi e 92.9%, con i ming he po en ial o ee-based ensemble me hods o iden i ying cogni i e
issues ea ly h ough sleep da a analysis.
The use o MRMR ea u e selec ion combined wi h Bagged T ees ("Iden i ica ion o C ucial Fac o s in Sleep Quali y")
achie ed 91.90%, unde lining he signi icance o li es yle p edic o s in sleep quali y analysis. Ano he high-pe o ming
model, he CRNN app oach ("A Hie a chical App oach o Diagnosis Using CRNN"), deli e ed 91.45% accu acy,
showcasing he capabili y o hyb id con olu ional- ecu en a chi ec u es in handling mul i-diso de and s age
classi ica ion p oblems.
Se e al s udies, including hose using Random Fo es (Gini Index and Tuned RF), main ained consis en accu acies
a ound 88%, alida ing he obus ness o adi ional ensemble classi ie s when applied o s anda dized da ase s like
Sleep Heal h and Li es yle. The BIRF senso -based s udy ("De ec ion and P edic ion o Sleep Diso de s by Co e BIRF
Senso s") also demons a ed a p ac ical applica ion wi h 88.6% sensi i i y and 89% speci ici y, p o ing e ec i e o
ea ly apnea de ec ion.
Meanwhile, deep lea ning-based models such as he 9-laye 1D-CNN showed p omise o mul i-class classi ica ion o
sleep s ages using PSG signals, al hough he exac accu acy was no speci ied. The hyb id TFCN model ocused on s age
classi ica ion using ISRUC and Sleep-EDFx da ase s achie ed Mac o-F1 sco es o 0.804 / 0.780, which, while lowe han
o he s, is no ewo hy gi en he complexi y o mul i-s age sleep classi ica ion. Las ly, he Ensemble SVM me hod did no
epo a speci ic accu acy bu was no ed o i s poo pe o mance when elying on combined models o e indi idual
classi ie s.
3. Resul s and analysis
The below Ba Cha isually ep esen s he accu acy o me ic pe o mance o se e al machine lea ning-based
esea ch s udies ocusing on sleep diso de classi ica ion o p edic ion. Among he lis ed s udies, Alshamma i’s
app oach, which u ilized a i icial neu al ne wo ks (ANNs) wi h gene ic algo i hm op imiza ion, achie ed he highes
accu acy o 92.92%, showcasing he supe io pe o mance o op imized neu al models in handling sleep da a. Nyholm
e al. closely ollowed wi h an accu acy o 92.90%, demons a ing he e ec i eness o G adien Boos ing algo i hms
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 2264-2270
2269
applied on he SNAC-B da ase o link sleep dis u bances wi h demen ia isks in olde adul s. The s udy by Wa unlawan
e al., which employed MRMR ea u e selec ion along wi h a Bagged T ees classi ie , a ained an imp essi e 91.90%
accu acy, highligh ing he impo ance o selec ing ele an ea u es in imp o ing model pe o mance. Simila ly,
Wadicha e al. u ilized a deep lea ning-based Con olu ional Recu en Neu al Ne wo k (CRNN) and achie ed 91.45%
accu acy in classi ying CAP sequences and sleep diso de s.
Figu e 1 Compa ison o Accu acy o Exis ing Algo i hms and Models
S udies based on ensemble me hods such as Random Fo es also showed obus esul s, wi h Zhang e al. achie ing
88.60% accu acy h ough he analysis o b ea hing pa e ns using RF senso s, and bo h Hidaya and Ta eq epo ing
88.00% accu acy using Random Fo es on he Sleep Heal h and Li es yle da ase , u he emphasizing he eliabili y o
ensemble lea ning o sleep classi ica ion asks. Las ly, Shao e al., despi e applying an ad anced hyb id in elligence
me hod combining TFCN and mul i- ask lea ning ac oss ISRUC and Sleep-EDFx da ase s, eco ded he lowes
pe o mance a 80.40% Mac o-F1 sco e, indica ing po en ial limi a ions o he applied deep lea ning app oach o
da ase complexi y. O e all, he cha illus a es ha while deep lea ning me hods can yield high pe o mance, simple
ensemble models like Random Fo es o well-op imized ANNs can pe o m equally o e en be e depending on he
da ase and p ep ocessing echniques used.
4. Conclusion
This p ojec op imizes machine lea ning echniques o he accu a e classi ica ion o sleep diso de s using he Sleep
Heal h and Li es yle Da ase . By applying ea u e selec ion, da a balancing, and ensemble me hods such as G adien
Boos ing and Vo ing classi ie s, he sys em enhances diagnos ic accu acy and eliabili y. The indings indica e ha
op imized models can signi ican ly imp o e sleep diso de de ec ion, o e ing a scalable and e icien solu ion o ea ly
diagnosis and pe sonalized heal h ecommenda ions. In he u u e, he sys em can be in eg a ed wi h wea able de ices
o eal- ime moni o ing, inco po a e mul i-modal da a sou ces such as EEG and ECG o enhanced classi ica ion, and
be de eloped in o a clinical decision suppo ool o heal hca e p o essionals. Addi ionally, deep lea ning app oaches
like CNNs and RNNs can be explo ed o cap u e complex sleep pa e ns, and he model can be deployed as a mobile o
web applica ion o wide accessibili y. These ad ancemen s will u he s eng hen he sys em's e ec i eness in ea ly
de ec ion, ea men planning, and o e all sleep heal h imp o emen .
Compliance wi h e hical s anda ds
Disclosu e o con lic o in e es
No con lic o in e es o be disclosed.
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 2264-2270
2270
Re e ences
[1] T. S. Alshamma i, ‘‘Applying machine lea ning algo i hms o he classi ica ion o sleep diso de s,’’ IEEE Access,
ol. 12, pp. 36110–36121, 2024.
[2] Y. Shao, B. Huang, L. Du, P. Wang, Z. Li, Z. Liu, L. Zhou, Y. Song, X. Chen, and Z. Fang, ‘‘Reliable au oma ic sleep s age
classi ica ion based on hyb id in elligence,’’ Compu . Biol. Med., ol. 173, May 2024, A . no. 108314.
[3] I. A. Hidaya , ‘‘Classi ica ion o sleep diso de s using andom o es on sleep heal h and li es yle da ase ,’’ J. Dinda:
Da a Sci., In . Technol., Da a Anal., ol. 3, no. 2, pp. 71–76, Aug. 2023
[4] W. Z. T. Ta eq, ‘‘Sleep diso de s de ec ion and classi ica ion using andom o es s algo i hm,’’ in Decision Making
in Heal hca e Sys ems. Cham, Swi ze land: Sp inge , 2024, pp. 257–266.
[5] M. Wa unlawan, P. Homsud, P. Sapphaphab, O. Rin hon, and S. Pechp asa n, ‘‘Iden i ica ion o c ucial ac o s in
sleep quali y using machine lea ning models and MRMR ea u e selec ion echnique,’’ in P oc. 15 h Biomed. Eng.
In . Con . (BMEiCON), Oc . 2023, pp. 1–5.
[6] Z. Zhang, T. B. Con oy, A. C. K iege , and E. C. Kan, ‘‘De ec ion and p edic ion o sleep diso de s by co e bed-
in eg a ed RF senso s,’’ IEEE T ans. Biomed. Eng., ol. 70, no. 4, pp. 1208–1218, Ap . 2023.
[7] A. Wadicha , S. Mu a ka, D. Shah, A. Bhu ane, M. Sha ma, H. S. Mi , and U. R. Acha ya, ‘‘A hie a chical app oach
o he diagnosis o sleep diso de s using con olu ional ecu en neu al ne wo k,’’ IEEE Access, ol. 11, pp.
125244–125255, 2023.
[8] J. Nyholm, A. N. Ghazi, S. N. Ghazi, and J. S. Be glund, ‘‘P edic ion o demen ia based on olde adul s’ sleep
dis u bances using machine lea ning,’’ Compu . Biol. Med., ol. 171, May 2024, A . no. 108126.
[9] S. K. Sa apa hy and D. Logana han, ‘‘Au oma ed classi ica ion o mul iclass sleep s ages classi ica ion using
polysomnog aphy signals: A ninelaye 1D-con olu ion neu al ne wo k app oach,’’ Mul imedia Tools Appl., ol.
82, no. 6, pp. 8049–8091, Ma . 2023.
[10] E. Alicko ic and A. Subasi, ‘‘Ensemble SVM me hod o au oma ic sleep s age classi ica ion,’’ IEEE T ans. Ins um.
Meas., ol. 67, no. 6, pp. 1258–1265, Jun. 2018
