Indonesian
J
our
nal
of
Electrical
Engineering
and
Computer
Science
V
ol.
41,
No.
2,
February
2026,
pp.
624
∼
632
ISSN:
2502-4752,
DOI:
10.11591/ijeecs.v41.i2.pp624-632
❒
624
A
new
h
ybrid
model
based
on
machine
lear
ning
and
fuzzy
logic
f
or
QoS
enhancing
in
IoT
Oussama
Lagnfdi,
Mar
ouane
Myyara,
Anouar
Darif
Laboratory
of
Inno
v
ation
in
Mathematics,
Applications,
and
Information
T
echnology
,
Polydisciplinary
F
aculty
,
Sultan
Moulay
Slimane
Uni
v
ersity
,
Beni
Mellal,
Morocco
Article
Inf
o
Article
history:
Recei
v
ed
Jul
22,
2025
Re
vised
Dec
16,
2025
Accepted
Jan
11,
2026
K
eyw
ords:
Fuzzy
logic
Genetic
algorithm
IoT
Machine
learning
algorithms
Quality
of
service
ABSTRA
CT
The
f
ast
e
xpansion
of
internet
of
things
(IoT)
de
vices
presents
a
more
compli-
cated
scenario
for
maintaining
a
stable
quality
of
service
(QoS),
which
w
ould
guarantee
the
netw
ork’
s
dependable
operation.
The
emer
gence
of
increasingly
comple
x
applications
that
call
for
additional
de
vices
mak
es
this
e
v
en
more
cru-
cial.
Adapti
v
e
intelligence
solutions
that
guarantee
optimal
netw
ork
beha
vior
are
therefore
required.
This
paper
presents
a
h
ybrid
optimized
solution
for
a
three-layer
IoT
netw
ork
that
models
the
application,
netw
ork,
and
perception
layers
of
an
IoT
netw
ork
using
m
achine
learning
and
fuzzy
logic
(FL).
This
method
guarantees
optimal
QoS
prediction
with
impro
v
ed
netw
ork
adaptability
by
using
fuzzy
membership
parameters.
When
the
number
of
de
vices
increases
from
100
to
1,500,
FLGA
maintains
an
a
v
erage
QoS
of
95%
to
87%,
while
FL
maintains
84%
and
RANDOM
maintains
79%.
At
the
applicat
ion
le
v
el,
genetic
algorithm
(GA)
continues
to
outperform
RANDOM
by
15.57%
and
FL
by
6.32%.
The
goal
of
this
paper
is
to
pro
vide
a
solid
netw
ork
solution
that
could
enhance
the
cons
istenc
y
of
QoS
performance
in
order
to
combat
the
increasingly
comple
x
scenario
of
an
IoT
netw
ork.
This
is
an
open
access
article
under
the
CC
BY
-SA
license
.
Corresponding
A
uthor:
Oussama
Lagnfdi
Laboratory
of
Inno
v
ation
in
Mathematics,
Applications,
and
Information
T
echnology
Polydisciplinary
F
aculty
,
Sultan
Moulay
Slimane
Uni
v
ersity
Beni
Mellal,
23000,
Morocco
Email:
lagnfdi.o@gmail.com
1.
INTR
ODUCTION
The
internet
of
things
(IoT)
is
a
rapidly
e
v
olving
technological
paradigm
b
uilt
on
interconnected
de-
vices—such
as
sensors,
smartphones,
and
radio-frequenc
y
identication
(RFID)
tags
that
communicate
via
the
Internet.
Ensuring
high
quality
of
service
(QoS)
is
essential
in
critical
application
domains
such
as
agriculture,
transportation,
healthcare,
and
manuf
acturing
[1],
[2].
Ho
we
v
er
,
maintaining
QoS
in
IoT
en
vironments
remains
a
signicant
challenge.
These
systems
operate
across
multiple
layers
perception
(sensors),
netw
ork,
and
appli-
cation
each
introducing
dist
inct
comple
xities
[3],
[4].
As
the
number
of
IoT
de
vices
increases,
ensuring
smooth
and
reliable
communication
becomes
increasingly
dif
cult.
K
e
y
challenges
include
heterogeneous
standards,
netw
ork
congestion,
and
signal
de
gradation,
all
of
which
can
impede
optimal
system
performance
[5].
Recent
studies
suggest
that
h
ybrid
metaheuristic
methods
typically
outperform
single-method
approaches
in
optimiz-
ing
IoT
system
performance
[6],
[7].
T
raditional
cloud-based
architectures,
where
computation
is
centralized,
often
f
ail
to
meet
the
stringent
real-time
requirements
of
delay-sensiti
v
e
applications.
Multi-access
edge
com-
puting
(MEC),
which
processes
data
closer
to
its
source,
mitig
ates
latenc
y
issues
[8],
maintaining
real-time
J
ournal
homepage:
http://ijeecs.iaescor
e
.com
Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian
J
Elec
Eng
&
Comp
Sci
ISSN:
2502-4752
❒
625
QoS
under
dynamic
netw
ork
conditions
is
challenging.
Dif
ferent
IoT
domains
ha
v
e
v
aried
requirements
;
for
e
xample,
smart
healthcare
and
urban
monitoring
prioritize
lo
w
netw
ork-layer
latenc
y
for
timely
critical
signals
[9],
[10],
while
smart
transportation
and
city
management
require
f
ast
data
processing
to
support
safety
,
emer
-
genc
y
response,
and
tr
af
c
control,
necessitating
a
holistic
QoS
strate
gy
[11]-[13].
Security-a
w
are
QoS
metrics
(e.g.,
encryption
o
v
erhead,
authentication
delay)
are
not
considered
here,
b
ut
the
proposed
fuzzy
frame
w
ork
could
accommodate
them
in
future
e
xtensions.
Fuzzy
logic
(FL)
and
metaheuristic
h
ybrids
sho
w
strong
potential
for
addressing
the
comple
xities
of
modern
IoT
en
vironments.
Approaches
such
as
fuzzy-based
multi-criteria
decision-making
and
metaheuristic
optimization
dynamically
balance
QoS
metrics,
including
e
x
ecution
time,
ener
gy
consumption,
and
communi-
cation
del
ays,
in
IoT
and
fog–cloud
sys
tems
[14],
[15].
Metaheuris
tic-based
techniques,
including
a
n
t
colon
y
optimization
and
impro
v
ed
seagull
optimization,
enable
adapt
i
v
e
task
scheduling
and
controller
placement
to
enhance
load
balancing
and
ener
gy
ef
cienc
y
across
heterogeneous
IoT
layers
[16],
[17].
In
wireless
sensor
netw
orks
(WSNs),
interv
al
T
ype-2
fuzzy
clustering
combined
with
heuristic
sleep
scheduling
e
xtends
netw
ork
lifetime
by
managing
uncertainties
in
node
ener
gy
le
v
els
and
uctuating
w
orkloads
[18],
[19].
Furthermore,
h
ybrid
metaheuristic
frame
w
orks,
such
as
the
combination
of
geneti
c
algorithms
(GA)
with
particle
sw
arm
op-
timization
(PSO),
impro
v
e
routing
reliability
and
throughput
in
dynamic
IoT
netw
orks
[20],
[21].
Inte
grating
softw
are-dened
netw
orking
(SDN)
with
heuristic
feature
selection
enhances
traf
c
cla
ssication
and
real-time
o
w
management
by
optimizing
the
placement
of
controllers
[22],
[23].
Finally
,
h
ybrid
fuzzy-metaheuristic
scheduling
methods
optimize
task
allocation,
reducing
latenc
y
and
maintaining
QoS
in
latenc
y-sensiti
v
e
edge
computing
scenarios
by
ef
fecti
v
ely
na
vig
ating
the
tr
ade-of
f
between
computational
cost
and
accurac
y
[24],
[25].
Despite
adv
ances,
most
e
xisting
methods
focus
on
single
QoS
metrics
o
r
indi
vidual
layers,
with
limited
attention
to
multi-layer
optimization
under
high
de
vice
density
,
highlighting
the
need
for
scalable
IoT
QoS
strate
gies.
T
o
address
this,
we
propose
a
holistic,
multi-layer
frame
w
ork
that
simultaneously
e
v
aluates
and
tunes
QoS
parameters
across
perception,
netw
ork,
and
application
layers.
By
inte
grating
FL
interpretability
with
GA-
based
membership
function
tuning,
the
frame
w
ork
impro
v
es
adaptability
and
reduces
root
mean
square
error
(RMSE)
under
v
arying
IoT
loads.
The
GA
adjusts
fuzzy
system
parameters
based
on
observ
ed
performance,
acting
as
a
learning
mechanism
that
enables
adaptation
to
comple
x
netw
ork
dynamics.
This
approach
addresses
high
de
vice
density
and
cross-layer
dependencies,
pro
viding
a
comprehensi
v
e
solution
for
end-to-end
QoS
enhancement
in
scalable
IoT
systems.
The
paper
is
or
g
anized
as
follo
ws:
section
2
presents
the
system
model,
section
3
details
the
proposed
approach,
section
4
presents
the
results,
and
section
5
concludes
the
study
.
2.
THE
PR
OPOSED
SYSTEM
MODEL
AND
PR
OBLEM
FORMULA
TION
The
IoT
architecture,
depicted
in
Figure
1,
is
structured
into
three
primary
layers.
The
perc
eption
layer
is
responsible
for
data
acquisition
from
ph
ysical
de
vices,
including
sensors,
RFID
tags,
and
actuators.
The
netw
ork
Layer
f
acilitates
data
transmission
through
v
arious
communication
protocols
such
as
W
i-Fi,
Ethernet,
ZigBee,
and
cellular
netw
orks
(4G/5G).
The
application
layer
processes
the
transmitted
data
to
deli
v
er
specic
services
to
end-users.
Figure
1.
IoT
three-layer
architecture
A
ne
w
hybrid
model
based
on
mac
hine
learning
and
fuzzy
lo
gic
for
QoS
enhancing
in
IoT
(Oussama
La
gnfdi)
Evaluation Warning : The document was created with Spire.PDF for Python.
626
❒
ISSN:
2502-4752
High
QoS
is
critical
in
a
three-l
ayer
IoT
system.
Each
layer
has
specic
requirements:
the
perception
layer
must
ensure
accurate
and
timely
data
acquisition;
the
netw
ork
layer
should
maintain
lo
w
latenc
y
and
minimal
pack
et
loss;
and
the
application
layer
must
deli
v
er
reliable,
scalable
services.
The
heterogeneous
and
dynamic
nature
of
IoT
en
vironments
introduces
uncertainty
,
resulting
in
a
comple
x
multi-objecti
v
e
optimiza-
tion
problem
under
stochastic
conditions.
T
o
address
this,
a
fuzzy
inference
system
aggre
g
ates
layer
-specic
metrics—such
as
latenc
y
,
throughput,
and
accurac
y—into
a
unied
QoS
score,
normalized
from
0%
to
100%.
QoS
=
P
i
x
i
·
µ
(
x
i
)
P
i
µ
(
x
i
)
(1)
Here,
x
i
denotes
possible
QoS
outcomes
and
µ
(
x
i
)
represents
the
de
gree
of
membership
for
each
outcome.
This
approach
consolidates
multiple
performance
metrics
into
a
single
score.
Maintaining
high
QoS
across
all
layers
therefore
requires
an
adapti
v
e
and
e
xible
opti
mization
frame
w
ork,
making
the
inte
gration
of
FL
with
GA
a
suitable
solution.
3.
METHOD
The
proposed
system,
illustrated
in
Figure
2,
is
designed
to
enhance
QoS
across
the
application,
netw
ork,
and
perception
layers
of
an
IoT
en
vironment.
Each
layer
is
e
v
aluated
using
a
dedicated
set
of
QoS
metrics
to
ensure
reliability
,
ef
cient
communication,
and
scalable
performance
under
increasing
de
vice
density
and
data
traf
c.
.
FCS
ML-A
Throughput
QoS App
FL
QoS App
Optimized
FCS
ML-A
QoS Net
FL
QoS Net
Optimized
FCS
ML-A
QoS Per
FL
QoS App
Optimized
Mean
QoS Total
Optimized
Network Layer
Application Layer
Perception Layer
Reliability
Latency
Packet Loss
Response
Time
Accurancy
Figure
2.
Multi-layer
QoS
optimization
model
The
o
v
erall
QoS
is
formulated
as
a
weighted
aggre
g
ation
of
the
application,
netw
ork,
and
perception
layer
QoS
v
alues,
Q
total
=
w
1
Q
app
+
w
2
Q
net
+
w
3
Q
perc
(2)
where
equal
importance
is
assumed
for
all
layers,
i.e.,
w
1
=
w
2
=
w
3
=
1
3
.
A
h
ybrid
fuzzy–genetic
optimiza-
tion
frame
w
ork
is
adopted,
in
which
the
GA
is
guided
by
the
RMSE,
to
impro
v
e
the
aggre
g
ated
QoS
across
all
IoT
layers.
RMSE
=
v
u
u
t
1
n
n
X
i
=1
(
y
i
−
ˆ
y
i
)
2
(3)
From
a
computational
standpoint,
let
P
represent
the
population
size,
G
the
number
of
generations,
n
the
number
of
training
samples,
and
R
the
number
of
fuzzy
rules
or
membership
function
parameters.
The
e
v
aluation
of
a
single
chromosome
requires
e
x
ecuting
the
fuzzy
inference
mechanism
o
v
er
all
n
samples,
leading
to
a
computational
comple
xity
of
O
(
n
×
R
)
.
As
the
GA
e
v
aluat
es
P
indi
viduals
in
each
generation,
the
Indonesian
J
Elec
Eng
&
Comp
Sci,
V
ol.
41,
No.
2,
February
2026:
624–632
Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian
J
Elec
Eng
&
Comp
Sci
ISSN:
2502-4752
❒
627
resulting
per
-generation
computational
cost
is
O
(
P
×
n
×
R
)
.
Consequently
,
after
G
generations,
the
o
v
erall
optimization
comple
xity
can
be
e
xpressed
as
O
(
P
×
G
×
n
×
R
)
.
In
practical
deplo
yments,
this
tuning
proce-
dure
is
performed
of
ine
or
at
sche
d
ul
ed
update
interv
als
on
g
ate
w
ay
or
serv
er
-le
v
el
nodes,
whereas
the
online
fuzzy
inference
process
incurs
an
approximately
constant
e
x
ecution
time
per
request.
As
a
result,
the
proposed
GA–fuzzy
frame
w
ork
remains
computationally
ef
cient
and
scalable
for
lar
ge-scale
IoT
systems.
Based
on
the
RMSE-based
tness
function
dened
in
(3),
a
genetic
algorithm
is
emplo
yed
to
tune
the
fuzzy
QoS
parameters.
The
complete
optimization
procedure
is
described
in
Algorithm
1.
Algorithm
1.
Genetic
algorithm
f
or
fuzzy
QoS
tuning
Requir
e:
IoT
Architecture,
N
=
50
,
G
=
100
,
P
c
=
0
.
8
,
P
m
=
0
.
01
Ensur
e:
Optimized
Fuzzy
P
arameters
(Best
Chromosome)
1:
Initialize
population
P
with
50
random
chromosomes
2:
f
or
generation
g
=
1
to
100
do
3:
Fitness:
Ev
aluate
R
M
S
E
for
each
C
i
∈
P
via
PureEdgeSim
4:
Calculate
Fitness(
C
i
)
←
1
/
(1
+
R
M
S
E
)
5:
Selection:
Perform
T
our
nament
Selection
for
elite
parents
6:
Cr
osso
v
er:
Apply
Multi-point
Cr
osso
v
er
(
P
c
=
0
.
8
)
7:
Mutation:
Mutate
of
fspring
chromosomes
(
P
m
=
0
.
01
)
8:
Update:
Replace
lo
w-performing
indi
viduals
with
of
fspring
9:
Retain
elite
chromosomes
to
maintain
population
inte
grity
10:
if
QoS
con
v
er
gence
reached
OR
g
=
100
then
11:
br
eak
and
identify
best
chromosome
12:
end
if
13:
end
f
or
14:
r
etur
n
Best
Chromosome
(Optimized
QoS
P
arameters)
4.
RESUL
TS
AND
DISCUSSION
The
GA–fuzzy
QoS
optimization
frame
w
ork
w
as
e
v
aluated
using
PureEdgeSim
[26]
in
a
t
hree-layer
IoT
architecture
with
a
high-density
scal
e
of
up
to
1,500
de
vices.
Layer
-wise
QoS
v
alues
were
optimized
via
a
GA
congured
with
a
population
size
of
50
and
100
generations.
T
o
ensure
rob
ust
global
search
while
maintaining
the
inte
grity
of
heuristic
rules,
we
utilized
a
crosso
v
er
probability
of
0.8
and
a
mutation
probability
of
0.01.
Selection
w
as
performed
via
tournament
selection.
These
parameters
were
specically
chosen
to
maximize
QoS
con
v
er
gence
in
comple
x,
high-density
scenarios,
where
local
optima
are
frequent.
T
ables
1
and
2
summarize
the
fuzzy
input
and
output
parameters
emplo
yed
to
assess
QoS
in
each
layer
.
T
able
1.
IoT
fuzzy
parameters
by
layer
Layer
Input
parameter
Fuzzy
sets
Range/UoD
App.
Throughput/Reliability
Lo
w
,
Med,
High
0–100
%
No.
of
de
vices
Fe
w
,
Mod,
Man
y
100–1500
Net.
Latenc
y/P
ack
et
loss
F
ast,
Med,
Slo
w
0–1
/
0–3
(s)
De
vice
load
Lo
w
,
Med,
High
0–100
%
Perc.
Accurac
y
Lo
w
,
Med,
High
0–100
%
Response
time
F
ast,
Med,
Slo
w
0–3
(s)
Acti
v
e
de
vices
Fe
w
,
Mod,
Man
y
100–1500
T
able
2.
Fuzzy
logic
output
parameters
P
arameters
Fuzzy
set
Range
(%)
App
QoS,
Nw
QoS,
Perc
QoS
Bad,
Medium,
Good
0–100
4.1.
A
pplication
lay
er
QoS
analysis
Figure
3
sho
ws
ho
w
QoS
changes
at
the
application
layer
as
the
number
of
IoT
de
vices
incre
ases,
while
comparing
the
dif
ferent
approaches.
The
GA
approach
consistently
achie
v
es
the
strongest
performance,
starting
at
nearly
95%
QoS
for
100
de
vices
and
slo
wly
decreasing
to
around
87%
when
the
number
reaches
1,500
de
vices
,
which
indicates
both
reliability
and
scalability
.
PSO
follo
ws
thi
s
trend,
maintaining
QoS
abo
v
e
90%
until
close
to
1,000
de
vi
ces,
after
which
a
sharper
decline
appears.
FL
be
gins
near
88%
b
ut
decreases
more
rapidly
as
the
de
vice
count
gro
ws,
whereas
the
RANDOM
approach
performs
the
w
orst,
starting
around
80%
and
dropping
qui
ckly
.
Ov
erall,
GA
deli
v
ers
the
most
stable
and
ef
fecti
v
e
QoS,
demon-
strating
better
adaptability
as
system
demands
continue
to
increase.
A
ne
w
hybrid
model
based
on
mac
hine
learning
and
fuzzy
lo
gic
for
QoS
enhancing
in
IoT
(Oussama
La
gnfdi)
Evaluation Warning : The document was created with Spire.PDF for Python.
628
❒
ISSN:
2502-4752
Figure
3.
Application
layer
QoS
for
dif
ferent
approaches
4.2.
Netw
ork
lay
er
QoS
analysis
Figure
4
illustrates
the
performance
of
each
method
in
maint
aining
netw
ork
layer
QoS
as
the
num-
ber
of
de
vices
increases.
GA
consistently
achie
v
es
QoS
abo
v
e
95%,
with
only
a
slight
decrease
un
de
r
higher
congestion,
demonstrating
rob
ust
performance
when
latenc
y
and
pack
et
loss
are
critical.
PSO
performs
com-
parably
,
maintaining
QoS
abo
v
e
90%,
indicating
ef
fecti
v
e
sw
arm-based
optimization,
though
slightly
less
re-
silient
than
GA
under
stress.
FL
e
xhibits
a
f
aster
decline
in
QoS,
suggesting
limited
adaptabil
ity
under
hea
vy
congestion.
RANDOM
sho
ws
signicant
uctuations
and
lacks
optimization.
Ov
erall,
adapti
v
e
e
v
olutionary
strate
gies
such
as
GA
and
PSO
outperform
both
static
FL
and
baseline
RANDOM
approaches
in
managing
netw
ork
comple
xity
.
Figure
4.
Netw
ork
layer
QoS
for
dif
ferent
approaches
4.3.
P
er
ception
lay
er
QoS
analysis
Figure
5
sho
ws
ho
w
each
approach
handles
QoS
at
the
perception
layer
,
looking
at
sensor
accurac
y
,
response
time,
and
data
consistenc
y
.
The
GA
approach
demonstrates
superior
performance,
maintaining
QoS
near
95%
with
only
a
slight
decline
as
de
vice
density
increases.
This
indicates
ef
fecti
v
e
tuning
of
the
fuzzy
membership
functions
for
sensory
data
conditions.
PSO
maintains
a
consistent
b
ut
lo
wer
QoS,
ranging
between
88%
and
90%.
In
contrast,
FL
e
xhibits
a
more
pronoun
c
ed
performance
de
gradation,
re
v
ealing
its
limited
adaptability
to
changing
sensing
conditions.
The
RANDOM
strate
gy
consistently
yields
the
lo
west
QoS,
high-
lighting
signicant
challenges
in
stability
and
scalability
.
Collec
ti
v
ely
,
these
results
underscore
the
adapti
v
e
rob
ustness
of
the
GA-based
optimization,
making
it
a
suitable
candidate
for
lar
ge-scale
and
unpredictable
IoT
en
vironments.
Indonesian
J
Elec
Eng
&
Comp
Sci,
V
ol.
41,
No.
2,
February
2026:
624–632
Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian
J
Elec
Eng
&
Comp
Sci
ISSN:
2502-4752
❒
629
Figure
5.
Perception
layer
QoS
for
dif
ferent
approaches
4.3.1.
Ov
erall
QoS
perf
ormance
T
able
3
sho
ws
the
combi
ned
QoS
for
all
three
IoT
layers,
where
GA
consistently
achie
v
es
the
highest
v
alues,
gradually
de
creasing
from
95%
to
87%
as
the
number
of
de
vices
increases
from
100
to
1,500.
PSO
ranks
second,
follo
wed
by
FL
and
RANDOM,
highlighting
GA
’
s
superior
performance
and
reliability
across
v
arying
system
sizes.
T
able
3.
QoS
performance
(%)
relati
v
e
to
the
number
of
IoT
de
vices
Algorithm
100
300
500
700
900
1100
1300
1500
RANDOM
87
85
84
83
82
81
80
79
FL
91
90
89
88
87
86
85
84
PSO
93
92
91
90
89
88
87
86
FL-GA
95
94
93
92
91
90
89
87
T
able
4
sho
ws
the
data
suggests
a
clear
trade-of
f:
by
allo
wing
a
longer
of
ine
optimization
period,
the
FL-GA
model
achie
v
es
a
15.57%
i
mpro
v
eme
nt
in
the
application
layer
and
a
6.03%
g
ain
in
the
Netw
ork
layer
compared
to
PSO.
Because
the
GA
is
more
resilient
to
local
optima
in
hi
gh-density
scenarios
(1,500
de
vices),
it
manages
the
stochastic
nature
of
the
perception
and
netw
ork
layers
more
ef
fecti
v
ely
.
Since
this
tuning
process
is
decoupled
from
real-time
operations,
the
superi
o
r
QoS
stability
pro
vided
by
the
GA
mak
es
it
the
most
rob
ust
solution
for
lar
ge-scale
IoT
deplo
yments
where
performance
quality
is
the
ultimate
metric
of
success.
T
able
4.
Percentage
impro
v
ement
of
FL-GA
vs.
Baselines
Comparison
App.
Net.
Perc.
T
otal
QoS
FL-GA
vs.
RANDOM
15.57%
12.57%
7.95%
9.64%
FL-GA
vs.
FL
6.32%
2.50%
2.97%
4.00%
FL-GA
vs.
PSO
3.33%
6.03%
6.38%
1.68%
4.4.
Computational
runtime
and
scalability
analysis
T
able
5
reports
The
mean
runtime
of
the
proposed
FL-GA
model
for
de
vice
dens
ities
up
to
1,500
de
vices
is
e
v
aluated
using
a
GA
(population
=
50,
generations
=
100).
This
runtime
corresponds
to
an
of
ine
training
phase
for
tuning
fuzzy
parameters
and
does
not
af
fect
online
system
operat
ion.
The
optimization
prioritizes
QoS
maximization
o
v
er
training
time,
as
reliability
is
critica
l
in
dense
IoT
en
vironments.
During
online
v
alidation,
RANDOM-based
tuning
results
in
the
highest
latenc
y
,
PSO
achie
v
es
lo
wer
delays
due
to
sw
arm-based
adaptation,
while
the
proposed
FL-GA
consistently
deli
v
ers
the
lo
west
latenc
y
o
wing
to
globally
optimized
fuzzy
parameters.
T
o
v
alidate
the
of
ine
optimization,
an
online
latenc
y
comparison
in
Figure
6
among
RANDOM,
FL,
PSO,
and
FL-GA
is
conducted,
where
FL-GA
consistently
achie
v
es
the
lo
west
delay
.
The
de
vice
scale
reects
realistic
dense
IoT
scenarios
within
PureEdgeSim
and
ensures
stable,
reproducible
e
v
aluation.
A
ne
w
hybrid
model
based
on
mac
hine
learning
and
fuzzy
lo
gic
for
QoS
enhancing
in
IoT
(Oussama
La
gnfdi)
Evaluation Warning : The document was created with Spire.PDF for Python.
630
❒
ISSN:
2502-4752
T
able
5.
Mean
runtime
performance
Algorithm
Mean
runtime
(Seconds)
Latenc
y
(s)
RANDOM
2,858.4
High
PSO
4,960.5
Lo
w
FL-GA
(Pr
oposed)
6,928.2
V
ery
lo
w
Figure
6
sho
w
latenc
y
v
alidation
results,
the
proposed
FL-GA
approach
consistently
outperforms
RANDOM,
FL,
and
PSO
across
a
ll
de
vice
densities
from
100
to
1,500
de
vices.
As
the
number
of
IoT
de
vices
increases,
all
methods
e
xperience
higher
latenc
y;
ho
we
v
er
,
FL-GA
maintains
the
lo
west
delay
,
increasing
from
0.89
s
to
1.63
s,
demonstrating
superior
scalability
.
This
impro
v
ement
is
attrib
uted
to
the
of
ine
genetic
tuning
of
fuzzy
parameters,
which
e
nables
more
ef
cient
decis
ion-making
during
online
e
x
ecution.
Compared
to
PSO
and
classical
FL,
FL-GA
achie
v
es
better
load
adaptation
under
dense
netw
ork
conditions,
v
alidating
the
ef
fecti
v
eness
of
the
of
ine
optimization
process.
The
simulation
results
demonstrate
the
reliability
and
ef
fecti
v
eness
of
the
GA–Fuzzy
approach
for
QoS
optimization
in
IoT
systems.
GA–Fuzzy
consistently
maintains
stable
performance
across
all
layers,
e
v
en
as
the
number
of
de
vices
gro
ws.
The
fuzzy
system
adapts
dynamically
,
ensuring
minimal
performance
de
gra-
dation
compared
with
PSO,
FL,
and
RANDOM
methods.
V
ari
ability
in
FL
and
RANDOM
highlights
the
challenges
of
static
or
non-adapti
v
e
approaches.
GA–Fuzzy
achie
v
es
high
QoS
while
requiring
careful
param-
eter
tuning
and
computational
resources,
indicating
that
it
pro
vides
a
scalable
and
rob
ust
solution,
particularly
suitable
for
lar
ge-scale
or
dynamic
IoT
deplo
yments.
Figure
6.
The
latenc
y
of
all
approaches
VS
the
number
of
IoT
de
vices
5.
CONCLUSION
This
paper
presents
a
multi-layer
h
ybrid
GA–FL
frame
w
ork
for
enhancing
QoS
across
the
perception,
netw
ork,
and
application
layers
of
IoT
systems.
By
inte
grating
the
interpretability
of
FL
with
the
adapti
v
e
optimization
capability
of
GA,
the
proposed
model
enables
ef
fecti
v
e
QoS
e
v
aluation
and
optimization
in
dense
IoT
en
vironments.
Simulation
results
obtained
using
PureEdgeSim
demonstrate
the
consistent
superiority
of
the
GA-based
approach.
When
the
number
of
IoT
de
vices
increases
to
1,500,
the
proposed
method
maintains
an
o
v
erall
QoS
between
95%
and
87%,
outperforming
both
classical
FL
and
RANDOM-based
strate
gies.
Sig-
nicant
impro
v
ements
are
also
observ
ed
at
the
application
layer
,
conrming
the
scalability
and
reliability
of
the
proposed
solution.
Although
the
e
v
aluation
is
limited
to
simulation-based
e
xperiments,
real-w
orld
f
actors
such
as
hardw
are
constraints,
protocol
o
v
erheads,
and
en
vironmental
interference
may
inuence
performance.
These
aspects
moti
v
ate
future
ef
forts
to
w
ard
real-w
orld
v
alidation,
e
xtended
scalability
analysis,
and
the
inte
gration
of
security-a
w
are
QoS
metrics
into
the
proposed
frame
w
ork.
A
CKNO
WLEDGMENTS
The
authors
w
ould
lik
e
to
thank
the
anon
ymous
re
vie
wers
for
their
v
aluable
comments.
Indonesian
J
Elec
Eng
&
Comp
Sci,
V
ol.
41,
No.
2,
February
2026:
624–632
Evaluation Warning : The document was created with Spire.PDF for Python.
Indonesian
J
Elec
Eng
&
Comp
Sci
ISSN:
2502-4752
❒
631
FUNDING
INFORMA
TION
This
research
recei
v
ed
no
e
xternal
funding.
A
UTHOR
CONTRIB
UTIONS
ST
A
TEMENT
The
CRediT
(Contrib
utor
Roles
T
axonomy)
authorship
contrib
ution
statement
for
this
study
is
sum-
marized
in
the
follo
wing
T
able,
which
details
the
specic
roles
and
responsibilities
of
each
author
.
Name
of
A
uthor
C
M
So
V
a
F
o
I
R
D
O
E
V
i
Su
P
Fu
Oussama
Lagnfdi
√
√
√
√
√
√
√
√
√
√
Marouane
Myyara
√
√
√
√
√
√
Anouar
Darifr
√
√
√
√
√
√
√
√
√
√
C
:
C
onceptualization
I
:
I
n
v
estig
ation
V
i
:
V
i
sualization
M
:
M
ethodology
R
:
R
esources
Su
:
Su
pervision
So
:
So
ftw
are
D
:
D
ata
Curation
P
:
P
roject
administration
V
a
:
V
a
lidation
O
:
Writing
-
O
riginal
Draft
Fu
:
Fu
nding
acquisition
F
o
:
F
o
rmal
analysis
E
:
Writing
-
Re
vie
w
&
E
diting
CONFLICT
OF
INTEREST
ST
A
TEMENT
The
authors
declare
that
the
y
ha
v
e
no
conict
of
interest.
D
A
T
A
A
V
AILABILITY
The
data
that
support
the
ndings
of
this
study
are
not
publicly
a
v
ailable
due
to
condenti
ality
restrictions.
REFERENCES
[1]
P
.
K
umar
,
R.
K
umar
,
G.
P
.
Gupta,
R.
T
ripathi,
A.
Jolf
aei,
and
A.
K.
M.
Najmul
Islam,
“
A
blockchain-orchestrated
deep
learning
approach
for
secure
data
transm
ission
in
IoT
-enabled
healthcare
system,
”
J
ournal
of
P
ar
allel
and
Distrib
uted
Computing
,
v
ol.
172,
pp.
69–83,
Feb
.
2023,
doi:
10.1016/j.jpdc.2022.10.002.
[2]
X.
Ma,
Z.
Dong,
W
.
Quan,
Y
.
Dong,
and
Y
.
T
an,
“Real-time
assessment
of
asphalt
pa
v
ement
moduli
and
traf
c
loads
using
monitor
-
ing
data
from
b
uilt-in
sensors
:
optimal
sensor
placement
and
identication
algorithm,
”
Mec
hanical
Systems
and
Signal
Pr
ocessing
,
v
ol.
187,
p.
109930,
Mar
.
2023,
doi:
10.1016/j.ymssp.2022.109930.
[3]
L.
N.
CheSuh,
R.
´
A.
Fern
´
andez-Diaz,
J.
M.
Alija-Perez,
C.
Bena
vides-Cuellar
,
and
H.
Alaiz-Moreton,
“Impro
v
e
quality
of
service
for
the
internet
of
things
using
blockchain
&machine
learning
algorithms,
”
Internet
of
Things
,
v
ol.
26,
p.
101123,
Jul.
2024,
doi:
10.1016/j.iot.2024.101123.
[4]
K.
C.
Serdaroglu,
S.
Baydere,
B.
Sao
v
apakhiran,
and
C.
Charnsripin
yo,
“Q-IoT
:
QoS-a
w
are
multilayer
service
architecture
for
multiclass
IoT
data
traf
c
management,
”
IEEE
Internet
of
Things
J
ournal
,
v
ol.
11,
no.
17,
pp.
28330–28340,
Sep.
2024,
doi:
10.1109/JIO
T
.2024.3402382.
[5]
Y
.
Liu,
J.
W
ang,
Z.
Y
an,
Z.
W
an,
and
R.
J
¨
antti,
“
A
surv
e
y
on
blockchain-based
trust
management
for
internet
of
things,
”
IEEE
Internet
of
Things
J
ournal
,
v
ol.
10,
no.
7,
pp.
5898–5922,
Apr
.
2023,
doi:
10.1109/JIO
T
.2023.3237893.
[6]
H.
K.
Apat,
B.
Sahoo,
V
.
Gosw
ami,
and
R.
K.
Barik,
“
A
h
ybrid
meta-heuristic
algorithm
for
multi-objecti
v
e
IoT
service
placement
in
fog
computing
en
vironments,
”
Decision
Analytics
J
ournal
,
v
ol.
10,
p.
100379,
Mar
.
2024,
doi:
10.1016/j.dajour
.2023.100379.
[7]
D.
Krek
o
vi
´
c,
P
.
Kri
vi
´
c,
I.
Podnar
ˇ
Zark
o,
M
.
K
u
ˇ
sek,
and
D.
Le-Phuoc,
“R
educing
communication
o
v
erhead
in
the
IoT–edge–cloud
continuum:
a
surv
e
y
on
protocols
and
data
reduction
strate
gies,
”
Internet
of
Things
,
v
ol.
31,
p.
101553,
May
2025,
doi:
10.1016/j.iot.2025.101553.
[8]
M.
M.
Hasan
et
al.
,
“The
journe
y
to
cloud
as
a
continuum:
opportunities,
challenges,
and
research
directions,
”
ICT
Expr
ess
,
v
ol.
11,
no.
4,
pp.
666–689,
Aug.
2025,
doi:
10.1016/j.icte.2025.04.015.
[9]
M.
Alaa,
A.
A.
Zaidan,
B.B.
Zaidan,
M.
T
alal,
and
M.L.M.
Kiah,
“
A
re
vie
w
of
smart
home
applications
based
on
internet
of
things,
”
J
ournal
of
Network
and
Computer
Applications
,
97,
48-65,
2017,
doi:
10.1016/j.jnca.2017.08.017.
[10]
A.
Rghioui,
J.
Lloret,
S.
Sendra,
and
A.
Oumnad,
“
A
smart
architecture
for
diabetic
patient
monitoring
using
machi
ne
learning
algorithms,
”
Healthcar
e
(Basel,
Switzerland),
v
ol.
8,
no.
3,
p.
348,
2020,
doi:
10.3390/healthcare8030348.
[11]
A.
Alnazir
,
R.
A.
Mokhtar
,
H.
Alhumyani,
E.
S.
Ali,
R.
A.
Saeed,
and
S.
Abdel-Khalek,
“Quality
of
services
based
on
intelligent
IoT
WLAN
MA
C
protocol
dynamic
real-time
applications
in
smart
cities,
”
Computational
intellig
ence
and
neur
oscience
,
v
ol.
2021,
2287531,
2021,
doi:
10.1155/2021/2287531.
[12]
S.
Abdellatif
and
L.
Ftouhi,
“Security-a
w
are
QoS
e
v
aluation
in
IoT
:
impacts
of
encryption
and
authentication,
”
IEEE
Internet
of
Things
J
ournal
,
v
ol.
8,
no.
7,
pp.
5652–5665,
2021.
A
ne
w
hybrid
model
based
on
mac
hine
learning
and
fuzzy
lo
gic
for
QoS
enhancing
in
IoT
(Oussama
La
gnfdi)
Evaluation Warning : The document was created with Spire.PDF for Python.
632
❒
ISSN:
2502-4752
[13]
H.
Qushtom
and
H.
R
abaya,
“Enhancing
the
QoS
of
IoT
netw
orks
with
lightweight
security
protocol
using
Contiki
OS,
”
Inter
-
national
J
ournal
of
Computer
Network
and
Information
Security
(IJC
NIS)
,
v
ol.
9,
no.
11,
pp.
27–35,
2017,
doi:
10.5815/ijc-
nis.2017.11.03.
[14]
S.
B.
Bose
and
S.
S.
Sujatha,
“Fuzzy
logic
based
QoS
management
and
monitoring
system
for
cloud
computing,
”
in
Pr
oceedings
3r
d
International
Confer
ence
Intellig
ent
Sustainable
Systems
(ICISS)
,
Dec.
2020,
doi:
10.1109/ICISS49785.2020.9315874.
[15]
V
.
Jain
and
B.
K
umar
,
“QoS-a
w
are
task
of
oading
in
fog
en
vironment
using
multi-agent
deep
reinfor
cement
learning,
”
J
ournal
of
Network
and
Systems
Mana
g
ement
,
2022,
doi:10.1007/s10922-022-09696-y
.
[16]
S.
K
umar
and
A.
Malik,
“Metaheuristic-based
optimized
and
resilient
controller
placement
in
SDN
enabled
IoT
netw
orks,
”
National
Academy
Science
Letter
s
,
No
v
.
2025,
doi:
10.1007/s40009-025-01860-7.
[17]
H.
M.
R.
Al-Khaf
aji,
“Impro
ving
quality
indicators
of
the
cloud-based
IoT
netw
orks
using
an
impro
v
ed
form
of
seagull
optimization
algorithm,
”
Futur
e
Internet
,
v
ol.
14,
no.
10,
art.
281,
Sep.
2022,
doi:10.3390/14100281.
[18]
A
.
Zaier
,
I.
Lahmar
,
M.
Y
ahia,
and
J.
Lloret,
“Interv
al
type-2
fuzzy
unequal
clustering
and
sleep
scheduling
for
IoT
-based
WSNs,
”
Ad
Hoc
Networks
,
v
ol.
175,
art.
103867,
2025,
doi:
10.1016/j.adhoc.2025.103867.
[19]
F
.
Oquendo,
“Fuzzy
architecture
description
for
handling
uncertainty
in
IoT
systems-of-systems,
”
in
Pr
oc.
15th
IEEE
International
Confer
ence
on
System
of
Systems
Engineering
(SoSE)
,
2020,
pp.
555–562,
doi:
10.1109/SoSE50414.2020.9130553.
[20]
P
.
Mishra,
N.
K
umar
,
and
W
.
Godfre
y
,
“
A
h
ybrid
meta-heuristic
solution
for
ener
gy-ef
cient
routing
in
softw
are-dened
wireless
sensor
netw
ork,
”
Resear
c
h
Squar
e
,
preprint,
2021,
doi:
10.21203/rs.3.rs-193270/v1.
[21]
A.
Satouf,
A.
Hamido
˘
glu,
¨
O.
M.
G
¨
ul,
A.
K
uusik,
L.
D.
Ata,
and
S.
Kadry
,
“Metaheuristic-based
task
scheduling
for
latenc
y-sensiti
v
e
IoT
applications
in
edge
computing,
”
Cluster
Computing
,
v
ol.
28,
no.
2,
art.
143,
2025,
doi:
10.1007/s10586-024-04878-6.
[22]
M.
A.
Aleisa,
“T
raf
c
classication
i
n
SDN-based
IoT
netw
ork
using
tw
o-le
v
el
fused
netw
ork
with
self-adapti
v
e
manta
ray
forag-
ing,
”
Scientic
Reports
,
v
ol.
15,
p.
881,
2025,
doi:
10.1038/s41598-024-84775-5.
[23]
S.
Sarohe,
S.
Harit,
and
M.
K
umar
,
“Load
balancing
in
SDN-IoT
netw
ork,
”
Lectur
e
Notes
in
Networks
and
Systems
,
v
ol.
23,
pp.
921–933,
2024,
doi:10.1007/978-981-97-7710-5
72.
[24]
J.
Aminu,
R.
Latip,
Z.
M.
Hana,
S.
Kamarudin,
and
D.
Gabi,
“Systematic
re
vie
w
of
metaheuristic-based
task
scheduling
strate
gies
in
edge
computing
en
vironments,
”
Disco
ver
Computing
,
v
ol.
28,
art.
307,
2025,
doi:10.1007/s10791-025-09819-4.
[25]
B.
A.
T
a
wfeeq,
A.
M.
Rahmani,
A.
K
oochari,
and
N.
J.
Na
vimipour
,
“Hybrid
optimization
method
for
social
internet
of
things
service
pro
vision
based
on
community
detection,
”
Engineering
Reports
,
2025.
[26]
A.
S.
Zamani,
R.
M.
Jag
adish,
B.
K
umar
,
A.
S.
Ghori,
and
S.
A.
Bh
yratae,
“Perspecti
v
es
of
m
achine
learning
in
the
con
v
er
gence
of
articial
intelligence
and
edge
computing,
”
i
n
Advances
in
AI
for
Cloud,
Edg
e
,
and
Mobile
Computing
Applications
,
Ne
w
Y
ork:
Apple
Academic
Press,
2025,
pp.
189–211.
BIOGRAPHIES
OF
A
UTHORS
Oussama
Lagnfdi
recei
v
ed
the
B.Sc.
de
gree
in
Ph
ysical
Matter
Science
in
2020
and
the
M.Sc.
de
gree
in
T
elecommunication
Systems
and
Computer
Netw
orks
in
2022
from
Sultan
Moulay
Slimane
Uni
v
ersity
,
Beni
Mellal,
Morocco.
He
is
currentl
y
pursuing
the
Ph.D.
de
gree
at
the
Labora-
toire
d’Inno
v
ation
en
Math
´
ematiques
et
Applications
et
T
echnologies
de
l’Information
(LIMA
TI),
Polydisciplinary
F
aculty
,
Sultan
Moulay
Slimane
Uni
v
ersity
.
His
research
interests
include
IoT
,
multi-access
edge
computing
(MEC),
cloud
computing,
AI,
deep
learning,
and
fuzzy
logic.
He
can
be
contacted
at
email:
lagnfdi.o@gmail.com.
Mar
ouane
Myyara
recei
v
ed
the
B.Sc.
de
gree
in
Electronic
and
T
elecommunication
Engineering
in
2019
and
the
M.Sc.
de
gree
in
T
elecommunication
Systems
and
Computer
Netw
orks
in
2021
from
Sulta
n
Moulay
Slimane
Uni
v
ersity
,
Beni
Mellal,
Morocco.
He
is
currently
pursuing
the
Ph.D.
de
gree
at
the
Laboratoire
d’I
nno
v
a
tion
en
Math
´
ematiques,
A
pplications
et
T
echnologies
de
l’Information,
Polydisciplinary
F
aculty
,
Sultan
Moulay
Slimane
Uni
v
ersity
.
His
research
interests
focus
on
MEC
netw
orks,
cloud
computing,
computation
of
oading,
and
the
IoT
.
He
can
be
contacted
at
email:
marouane.myyara@usms.ac.ma.
Anouar
Darif
recei
v
ed
the
B.Sc.
de
gree
in
Informatics,
Electrical
Engineering,
Elec-
tronics,
and
Automation
(IEEA)
from
Dhar
El
Mahraz
F
aculty
of
Sciences,
Mohamed
Ben
Abdellah
Uni
v
ersity
,
Fez,
Morocco,
in
2005,
the
Dipl
ˆ
ome
d’
´
Etudes
Sup
´
erieures
Approfondies
(DESA)
in
Com-
puter
Science
and
T
elecommunications
from
the
F
aculty
of
Sciences,
Rabat,
Morocco,
in
2007,
and
the
Ph.D.
de
gree
in
Computer
Science
and
T
elecommunications
from
the
F
aculty
of
Sci
ences,
Rabat,
in
2015.
He
is
currently
a
Research
and
T
eaching
Associate
at
the
Polydisciplinary
F
aculty
,
Sultan
Moulay
Slimane
Uni
v
ersity
,
Beni
Mellal,
Morocco.
His
research
interests
include
WSNs,
MEC,
IoT
,
cloud
computing,
and
neural
netw
orks.
He
serv
es
as
a
re
vie
wer
for
se
v
eral
international
journals
and
conferences.
He
can
be
contacted
at
email:
anouar
.darif@gmail.com.
Indonesian
J
Elec
Eng
&
Comp
Sci,
V
ol.
41,
No.
2,
February
2026:
624–632
Evaluation Warning : The document was created with Spire.PDF for Python.