IAES
Inter
national
J
our
nal
of
Articial
Intelligence
(IJ-AI)
V
ol.
14,
No.
3,
June
2025,
pp.
1951
∼
1959
ISSN:
2252-8938,
DOI:
10.11591/ijai.v14.i3.pp1951-1959
❒
1951
Comparati
v
e
analysis
of
machine
lear
ning
models
f
or
fak
e
news
detection
in
social
media
Bahaa
Eddine
Elbaghazaoui
1,2
,
Mohamed
Amnai
2
,
Y
oussef
F
akhri
2
,
Ali
Choukri
2
,
Nor
eddine
Gherabi
3
1
National
School
of
Applied
Sciences
of
Sultan
Moulay
Slimane
Uni
v
ersity
,
Beni
Mellal,
Morocco
2
Laboratory
of
Computer
Sciences
Research,
F
aculty
of
Sciences,
Ibn
T
of
ail
Uni
v
ersity
K
enitra,
K
enitra,
Morocco
3
National
School
of
Applied
Sciences
of
Sultan
Moulay
Slimane
Uni
v
ersity
,
Khouribg
a,
Morocco
Article
Inf
o
Article
history:
Recei
v
ed
Sep
2,
2024
Re
vised
No
v
22,
2024
Accepted
Jan
27,
2025
K
eyw
ords:
Data
proling
F
ak
e
ne
ws
Information
dif
fusion
Machine
learning
Social
media
ABSTRA
CT
The
rapid
rise
of
information
sharing
on
social
media
has
amplied
the
spread
of
f
ak
e
ne
ws,
making
its
detection
increasingly
critical.
As
f
ak
e
ne
ws
continues
to
proliferate,
the
need
for
ef
cient
detection
mechanisms
has
become
more
ur
gent
to
protect
users
from
misinformation
and
disinformation.
This
paper
presents
a
comparati
v
e
analysis
of
multiple
machine
learning
models
for
detecting
te
xt-
based
f
ak
e
ne
ws
on
social
media
platforms.
Using
mode
ls
such
as
gradient
boosting,
XGBoost,
and
linear
support
v
ector
classier
(SVC)
on
the
Infor
-
mation
Security
and
Object
T
echnology
(ISO
T)
f
ak
e
ne
ws
dataset,
the
study
demonstrates
that
gradient
boosting
achie
v
es
the
highest
accurac
y
of
99.61%,
while
XGBoost
pro
vides
a
strong
balance
with
99.59%
accurac
y
and
a
signi-
cantly
lo
wer
e
x
ecution
time,
making
it
more
suitable
for
real-time
applications.
These
results
of
fer
v
aluable
insights
into
the
trade-of
fs
between
accurac
y
and
computational
ef
cienc
y
,
contrib
uting
to
the
de
v
elopment
of
more
practical
de-
tection
systems
and
future
research
in
the
eld.
This
is
an
open
access
article
under
the
CC
BY
-SA
license
.
Corresponding
A
uthor:
Bahaa
Eddine
Elbaghazaoui
National
School
of
Applied
Sciences
in
Beni
Mellal,
Sultan
Moulay
Slimane
Uni
v
ersity
Beni
Mellal,
Morocco
Email:
b
.elbaghazaoui@usms.ac.ma
1.
INTR
ODUCTION
Social
media
users
frequently
share,
lik
e,
or
repost
content,
which
leads
to
the
widespread
dis
sem-
ination
of
information,
including
ne
ws
from
independent
authors
[1].
This
uncontrolled
o
w
of
information
has
contrib
uted
to
the
rise
of
f
ak
e
ne
ws,
a
gro
wing
global
issue
[2].
While
the
concept
of
f
ak
e
ne
ws
is
not
ne
w
,
the
perv
asi
v
e
use
of
social
media
has
signicantly
amplied
its
impact,
f
acilitating
the
rapid
s
pread
of
misinformation
[3].
The
ur
genc
y
of
addressing
f
ak
e
ne
ws
lies
in
its
potential
to
harm
both
indi
viduals
and
society
[4].
F
ak
e
ne
ws,
dened
as
intentionally
misleading
information
disseminated
through
ne
ws
outlets
or
social
media,
can
inuence
elections,
deepen
political
di
vides,
and
shape
public
perception
in
dangerous
w
ays
[5].
The
perv
asi
v
e
spread
of
f
ak
e
ne
ws
undermines
the
truth
within
the
information
ecosystem
and
misleads
readers
into
accepting
biased
or
inaccurate
information,
t
hereby
af
fecting
ho
w
people
interpret
real
ne
ws
[6].
In
response
to
this
gro
wing
problem,
this
paper
addresses
the
challenge
of
detecting
f
ak
e
ne
ws
by
applying
adv
anced
machine
learning
algorithms
to
analyze
and
classify
f
alse
information
shared
on
social
media
platforms.
Identifying
f
ak
e
ne
ws
accurately
is
essential
for
preserving
the
inte
grity
of
online
discourse
and
protecting
users
from
misinformation.
J
ournal
homepage:
http://ijai.iaescor
e
.com
Evaluation Warning : The document was created with Spire.PDF for Python.
1952
❒
ISSN:
2252-8938
T
raditional
computational
methods,
such
as
relying
on
satirical
ne
ws
sources
or
f
act-checking
web-
sites,
ha
v
e
limitations,
including
a
narro
w
focus
on
specic
topics
lik
e
politics
and
the
requirement
for
human
e
xpertise
[7].
These
constraints
hinder
the
de
v
elopment
of
models
that
can
generalize
across
multiple
domains,
which
is
necessary
for
an
ef
fecti
v
e
and
scalable
f
ak
e
ne
ws
detection
system.
The
primary
objecti
v
e
of
this
study
is
to
identify
the
most
ef
fecti
v
e
machine
learning
model
for
detecting
f
ak
e
ne
ws
in
social
media
posts.
W
e
achie
v
e
this
by
collecting
and
cleaning
a
dataset,
applying
v
arious
machine
learning
models,
and
comparing
their
performance
to
determine
the
optimal
approach
for
f
ak
e
ne
ws
detection.
This
study
contrib
utes
to
the
eld
of
f
ak
e
ne
ws
detection
by
pro
viding
a
comparati
v
e
analysis
of
multiple
machine
learning
models,
specically
gradient
boosting,
XGBoost,
and
linear
linear
support
v
ector
classier
(SVC),
applied
to
a
lar
ge
social
media
dataset.
Unlik
e
pre
vious
research
that
often
focuses
on
a
single
detection
method
or
narro
w
use
cases,
this
study
e
v
aluates
both
the
accurac
y
and
e
x
ecution
time
of
v
arious
models,
of
fering
insights
into
their
trade-of
fs
in
real-w
orld
applica
tions.
Additionally
,
the
study
demonstrates
the
practical
rele
v
ance
of
balancing
computational
ef
cienc
y
with
detection
precision,
making
it
a
signicant
contrib
ution
for
the
de
v
elopment
of
scalable,
real-time
f
ak
e
ne
ws
detection
systems.
The
ndings
pro
vide
a
v
aluable
foundation
for
future
research
in
impro
ving
the
detection
of
misinformation,
with
a
particular
emphasis
on
real-time
detection
capabilities
and
broader
generalization
across
domains.
This
paper
is
structured
as
follo
ws:
section
2
re
vie
ws
the
e
xisting
literature
on
f
ak
e
ne
ws
detection,
highlighting
g
aps
in
current
approaches
and
outlining
the
objecti
v
es
and
methodology
of
this
study
.
Section
3
e
xamines
the
spread
of
misinformation
on
social
media
and
e
xplains
ho
w
it
can
be
proled
using
data,
including
details
on
data
collection,
cleaning,
and
preprocessing
techniques.
Section
4
describes
the
application
of
v
arious
machine
learning
models
for
detecting
f
ak
e
ne
ws,
and
compares
their
performance
based
on
accurac
y
and
e
x
ecution
time.
Finally
,
section
5
e
v
aluates
the
results,
interprets
the
ndings
in
comparison
to
pre
vious
studies,
discusses
the
limitations
of
the
study
,
and
of
fers
suggestions
for
future
research,
including
the
potential
for
real-time
detection
systems
and
the
inte
gration
of
multimedia
data.
2.
RELA
TED
W
ORKS
This
section
outlines
signicant
early
research
on
the
concept
of
f
ak
e
ne
ws
and
pro
vides
an
o
v
ervie
w
of
the
gro
wing
body
of
studies
e
xamining
its
dissemination
[8].
The
adv
ent
of
the
w
orld
wide
web
(WWW)
and
the
rapid
adoption
of
social
media
platforms
lik
e
F
acebook
and
T
witter
ha
v
e
re
v
olutionized
the
w
ay
information
is
shared,
enabling
unprecedented
le
v
els
of
content
dissemination
[9].
Social
media
has
allo
wed
consumers
to
produce
and
share
more
information
than
e
v
er
before,
some
of
which
is
inaccurate
or
misleading,
contrib
uting
signicantly
to
the
spread
of
f
ak
e
ne
w
[10].
Scholars
ha
v
e
conceptualized
f
ak
e
ne
ws
in
v
arious
w
ays,
b
ut
the
core
denition
remains
consi
stent
across
dif
ferent
interpretations
[11].
F
ak
e
ne
ws
typically
refers
to
content
that
is
deliberately
created
and
dis-
seminated
with
the
intention
of
decei
ving
people
into
belie
ving
f
alsehoods
[12].
This
includes
myths,
rumors,
conspirac
y
theories,
hoax
es,
and
other
forms
of
misleading
or
incorrect
information,
which
are
frequently
shared
on
social
media
platforms
either
intentionally
or
inadv
ertently
[13].
A
pertinent
e
xample
of
the
impact
of
f
ak
e
ne
ws
i
s
the
use
of
articial
intelligence
(AI)-generated
content
in
v
olving
prominent
gures
lik
e
Elon
Musk
[14].
F
or
instance,
a
f
abricated
image
circulated
online
depicts
Musk
endorsing
a
nancial
scheme
called
”Quantum
AI,
”
f
alsely
claiming
that
it
allo
ws
Australians
to
“get
rich
quick.
”
In
reality
,
the
follo
wing
Figure
1,
designed
to
mimic
a
CNN
ne
ws
broadcast,
is
part
of
a
broader
scam
that
e
xploit
s
AI
to
decei
v
e
vie
wers
into
in
v
esting
in
a
fraudulent
v
enture.
This
case
illustrates
ho
w
sophisticated
technologies,
such
as
AI,
can
be
manipulated
to
create
con
vincing
f
ak
e
content,
complicating
ef
forts
to
discern
truth
from
deception
on
social
media
platforms.
The
use
of
well-kno
wn
gures
lik
e
Elon
Musk,
combined
with
credible-looking
ne
ws
formats,
can
easily
trick
unsuspecting
vie
wers
into
belie
ving
f
alse
claims.
This
underscores
the
im
portance
of
media
literac
y
and
the
ability
to
critically
assess
the
authenticit
y
of
online
content.
Moreo
v
er
,
the
CO
VID-19
pandemic
has
heightened
the
rele
v
ance
of
f
ak
e
ne
ws
research,
with
studies
e
xploring
the
link
between
social
media
and
the
spread
of
misinformation
related
to
public
health
[15].
Dur
-
ing
the
pandemic,
f
ak
e
ne
ws
about
the
corona
virus,
such
as
e
xaggerated
death
tolls
and
unfounded
conspirac
y
theories,
spread
rapidly
on
social
media,
posing
signicant
challenges
to
public
health
ef
forts
[16].
This
mis-
information
has
fueled
public
anxiety
,
created
confusion,
and
hindered
the
ef
forts
of
healthcare
w
ork
ers
and
public
health
systems
in
managing
the
crisis
ef
fecti
v
ely
.
Int
J
Artif
Intell,
V
ol.
14,
No.
3,
June
2025:
1951–1959
Evaluation Warning : The document was created with Spire.PDF for Python.
Int
J
Artif
Intell
ISSN:
2252-8938
❒
1953
The
primary
objecti
v
e
of
this
article
is
to
de
v
elop
methods
for
identifying
f
ak
e
ne
ws
across
v
arious
databases.
By
le
v
eraging
appropriate
machine
learning
models,
we
aim
to
accurately
predict
and
detect
f
ak
e
ne
ws
and
misinformation.
The
outcomes
of
this
study
will
aid
in
identifying
f
alse
information
and
help
users
a
v
oid
being
misled
by
decepti
v
e
content,
thereby
reducing
the
impact
of
f
ak
e
ne
ws
on
public
discourse.
While
pre
vious
studies
ha
v
e
focused
on
either
indi
vidual
machine
learning
models
or
specic
doma
ins
such
as
politics
or
public
health,
the
no
v
elty
of
this
w
ork
lies
in
its
comprehensi
v
e
comparati
v
e
analysis
of
mul-
tiple
machine
learning
models
for
detecting
f
ak
e
ne
ws
acr
o
s
s
a
broader
domain.
Unlik
e
e
xisting
research
that
often
emphasizes
accurac
y
alone,
our
study
e
v
aluates
both
t
he
accurac
y
and
e
x
ecution
time
of
models
lik
e
gradient
boosting,
XGBoost,
and
linear
SVC,
making
it
one
of
the
fe
w
w
orks
to
consider
computational
ef
cienc
y
alongside
detection
performance.
This
dual
focus
on
precision
and
speed
is
critical
for
de
v
eloping
real-time
f
ak
e
ne
ws
detection
systems
that
can
be
applied
across
di
v
erse
datasets
and
social
media
platforms.
Furthermore,
by
e
xploring
the
trade-of
fs
between
model
ef
fecti
v
eness
and
resource
ef
cienc
y
,
our
w
ork
of
fers
practical
insights
that
can
be
applied
to
the
real-w
orld
deplo
yment
of
scalable
detection
systems.
Figure
1.
AI-generated
f
ak
e
CNN
broadcast
featuring
Elon
Musk
promoting
fraudulent
’Quantum
AI’
scheme
3.
CHALLENGES
IN
F
AKE
NEWS
DETECTION
Once
ne
ws
is
released,
a
lar
ge
number
of
users
may
eng
age
in
its
spread
across
online
social
media
netw
orks,
where
users
can
share
their
thoughts
and
opinions
[17].
F
or
each
ne
ws
item,
numerous
posts
can
be
observ
ed
and
collected
on
these
platforms.
Ho
we
v
er
,
man
y
of
these
un
v
eried
posts
are
f
alse
[18].
Pre
vious
attempts
to
utilize
ne
ws
propag
ation
for
f
ak
e
ne
ws
detection
ha
v
e
emplo
yed
v
arious
feature
s.
Most
research
in
thi
s
area
has
focused
on
supervised
learning
approaches,
while
fe
wer
studies
ha
v
e
e
xplored
unsupervised
models.
Additionally
,
man
y
of
these
methods
rely
on
a
single
machine
learning
model
to
b
uild
a
special-
ized
frame
w
ork
for
detection.
In
contrast,
this
paper
aims
to
tackle
the
c
hallenge
of
f
ak
e
ne
ws
detection
by
e
v
aluating
multiple
models
to
identify
the
one
that
of
fers
the
best
balance
of
speed
and
accurac
y
.
4.
SPEADING
F
AKE
NEWS
F
ak
e
ne
ws
is
disseminated
through
social
and
mainstream
electronic
media
in
v
arious
forms,
i
ncluding
comments,
political
agendas,
ne
ws
articles,
rumors,
and
satire.
It
is
widely
used
to
spread
misleading
informa-
tion,
f
alse
persuasion,
and
confusion,
posing
a
signicant
t
hreat
to
public
trust
in
online
acti
vities
such
as
social
interactions,
e-commerce,
and
media
consumption
[19].
The
challenge
of
detecting
and
mitig
ating
f
ak
e
ne
ws
is
comple
x
due
to
its
dynamic
and
heterogeneous
nature.
Researchers
in
natural
language
processing
(NLP)
ha
v
e
de
v
eloped
numerous
solutions
to
address
this
issue
[20].
The
k
e
y
challenges
f
aced
by
academic
researchers
in
this
eld
include
the
di
v
ersity
,
speed,
quantity
,
and
persistence
of
f
ak
e
ne
ws,
as
illustrated
in
Figure
2.
F
ak
e
ne
ws
often
starts
as
disinformation,
which
is
deliberately
f
abricated
for
a
specic
purpose,
and
later
spreads
as
misinformation,
when
people
unkno
wingly
share
incorrect
information.
The
primary
moti
v
a-
tions
behind
the
creation
of
f
ak
e
ne
ws
include:
–
Misinterpretation
or
distortion
of
true
ne
ws
–
Prot
generation
through
clicks
or
website
traf
c
–
Promotion
of
indi
viduals,
political
parties,
or
specic
vie
wpoints
–
Misunderstandings
from
jok
es
or
parody
content
being
tak
en
seriously
Due
to
conrmation
bias,
people
are
more
lik
ely
to
accept
information
as
true
if
it
aligns
with
their
e
xisting
beliefs
[21].
Ev
en
those
who
do
not
typically
belie
v
e
in
such
content
may
be
inuenced
by
its
shocking
nature,
leading
them
to
share
it.
Compar
ative
analysis
of
mac
hine
learning
models
for
...
(Bahaa
Eddine
Elba
ghazaoui)
Evaluation Warning : The document was created with Spire.PDF for Python.
1954
❒
ISSN:
2252-8938
Figure
2.
Core
research
areas
in
f
ak
e
ne
ws
studies
5.
D
A
T
A
PR
OFILING
AND
F
AKE
NEWS
The
core
b
usiness
of
man
y
technology
companies
in
v
olv
es
creating
identity
systems
based
on
data
collection
and
proling
[22].
This
has
signicant
implications
for
ne
ws
or
g
anizations,
which
often
struggle
to
deli
v
er
content
ef
fecti
v
ely
on
platforms
lik
e
F
acebook
and
T
witter
[23].
These
technology
platforms
ha
v
e
the
po
wer
to
shape
audiences
and
inuence
their
demands,
holding
much
of
the
data
necessary
to
fully
understand
ho
w
informati
on
reaches
i
ts
intended
audience.
This
situa
tion
creates
modern
data
blind
spots
and
contrib
utes
to
a
gro
wing
mistrust
between
ne
ws
or
g
anizations
and
the
public.
W
ithout
accurate
and
comprehensi
v
e
com-
munication
of
vital
f
acts,
the
reality
of
information
can
become
distorted.
Data
proling
is
crucial
in
understanding
the
v
alue
of
information
and
identifying
tools
that
can
detect
f
ak
e
ne
ws
on
social
media
[24].
It
supports
the
de
v
elopment
of
systems
that
automatically
detect
misinforma-
tion
in
ne
ws.
Metadata
feat
u
r
es
such
as
account
names,
images,
follo
wers,
number
of
lik
es,
tags,
and
release
dates
can
pro
vide
signals
that
indicate
suspicious
acti
vities
and
help
identify
f
ak
e
ne
ws.
The
internet
hosts
data
in
v
arious
formats,
including
te
xt,
images,
video,
and
audio.
Detecting
and
classifying
unstructured
ne
ws
published
online
can
be
challenging,
often
requiring
human
e
xpertise.
Ho
we
v
er
,
computational
techniques,
including
data
proling,
can
be
emplo
yed
to
detect
anomalies
that
dif
ferentiate
decepti
v
e
te
xt
articles
from
f
ac
t-based
ones.
Additionally
,
data
proling
can
be
used
to
analyze
the
spread
of
f
ak
e
ne
ws
compared
to
real
ne
ws,
pro
viding
insights
into
ho
w
the
dissemination
of
f
alse
information
on
the
internet
dif
fers
from
that
of
accurate
reporting.
6.
MA
CHINE
LEARNING
AND
F
AKE
NEWS
AI,
machine
learning,
and
data
science
are
v
aluable
tools
for
addressing
interdisciplinary
problems,
including
the
chal
lenge
of
f
ak
e
ne
ws
[25].
These
elds
are
often
distrib
uted
across
multiple
sub-disciplines
within
the
data
analysis
community
.
The
spread
of
f
ak
e
ne
ws
has
led
to
widespread
distrus
t
in
media,
politics,
and
established
institutions
globally
.
While
emer
ging
technologies
lik
e
AI
ha
v
e
the
potential
to
e
xacerbate
the
problem,
the
y
can
also
be
harnessed
to
combat
misinformation.
F
or
e
xample
,
deepf
ak
e
technology
,
which
uses
AI
to
manipulate
audio,
images,
and
videos,
can
cause
indi
viduals
to
appear
to
say
or
do
things
the
y
ne
v
er
actually
did
[26].
This
inno
v
ation
could
usher
in
a
ne
w
w
a
v
e
of
misinformation,
easily
distrib
uted
via
social
media.
The
creation
and
dissem
ination
of
f
alse
information
are
often
dri
v
en
by
economic
and
ideological
moti
v
es
[27].
Unfortunately
,
ef
forts
to
combat
f
ak
e
ne
ws
are
unlik
ely
to
deter
t
hose
with
v
ested
interests
from
de
v
eloping
ne
w
methods
to
spread
misinformation
[28].
Research
in
this
area
often
focuses
on
understanding
wh
y
people
are
susceptible
to
f
alse
information,
as
these
vulnerabilities
are
frequently
e
xploited
by
those
who
create
and
spread
f
ak
e
ne
ws.
Identifying
f
ak
e
ne
ws
requires
a
deep
understanding
of
the
issue,
which
is
challenging
because
man
y
people
tend
to
belie
v
e
misinformation
without
v
erifying
it,
simply
becaus
e
the
y
are
not
a
w
are
of
the
full
conte
xt.
Numerous
studies
ha
v
e
focused
on
identifying
and
cate
gorizing
f
ak
e
ne
ws
on
social
media
platforms
such
as
F
acebook
and
T
witter
.
F
ak
e
ne
ws
is
conceptually
classied
into
v
arious
forms,
and
this
clas
sication
is
then
used
to
de
v
elop
machine
learning
models
that
can
generalize
across
dif
ferent
domains.
Int
J
Artif
Intell,
V
ol.
14,
No.
3,
June
2025:
1951–1959
Evaluation Warning : The document was created with Spire.PDF for Python.
Int
J
Artif
Intell
ISSN:
2252-8938
❒
1955
Machine
learning
of
fers
a
po
werful
and
automated
w
ay
to
detect
f
ak
e
ne
ws.
When
f
ak
e
ne
ws
i
s
posted,
a
machine
learning
system
can
analyze
the
content
and
determine
its
authenticity
.
Researchers
are
acti
v
ely
w
orking
on
de
v
eloping
the
most
ef
fecti
v
e
machine
learning
classiers
to
identify
f
ak
e
ne
ws.
The
accurac
y
of
these
classiers
is
crucial;
if
a
classier
f
ails
to
correctly
identify
f
alse
ne
ws,
it
could
ha
v
e
harmful
consequences
for
man
y
indi
viduals.
The
accurac
y
of
a
classier
depends
on
the
quality
of
its
training.
A
well-
trained
model
can
achie
v
e
higher
accurac
y
in
detecting
f
ak
e
ne
ws,
and
v
arious
machine
learning
classiers
can
be
emplo
yed
for
this
purpose.
7.
EXPERIMENT
AL
EV
ALU
A
TION
In
our
study
,
we
utilized
the
ISO
T
F
ak
e
Ne
ws
Dataset,
which
w
as
sourced
from
Kaggle.
This
d
a
taset
is
widely
used
for
f
ak
e
ne
ws
detection
research
and
cont
ains
a
comprehensi
v
e
collection
of
real
and
f
ak
e
ne
ws
articles.
The
ISO
T
F
ak
e
Ne
ws
Dataset,
created
by
the
Information
Security
and
Object
T
echnology
(ISO
T)
research
group
at
the
Uni
v
ersity
of
V
ictoria,
is
widely
used
for
f
ak
e
ne
ws
detection
tasks
[29].
The
dataset
is
structured
into
tw
o
cate
gories:
real
ne
ws
and
f
ak
e
ne
ws.
The
follo
wing
are
the
k
e
y
statistics:
a)
T
otal
number
of
articles:
the
dataset
contains
a
total
of
44,898
articles.
b)
Real
ne
ws:
21,417
real
ne
ws
articles
were
collected
from
credible
ne
ws
sources
such
as
Reuters.
c)
F
ak
e
ne
ws:
23,481
f
ak
e
ne
ws
articles
were
g
athered
from
unreliable
websites
agged
by
Politif
act
and
W
ikipedia.
d)
T
ime
period:
the
dataset
co
v
ers
articles
published
mainly
in
2016
and
2017,
particularly
during
signi-
cant
political
e
v
ents.
e)
Data
format:
the
dataset
is
pro
vided
in
CSV
format
with
the
follo
wing
columns:
–
T
itle:
the
title
or
headline
of
the
article.
–
T
e
xt:
the
full
te
xt
or
body
content
of
the
article.
–
Label:
a
binary
label
indicating
whether
the
article
is
real
(1)
or
f
ak
e
(0).
This
dataset
is
frequently
used
for
training
and
e
v
aluating
machine
learning
models
to
detect
f
ak
e
ne
ws
[30].
Common
metrics
for
e
v
aluation
include
accurac
y
,
precision,
recall,
and
F1-score.
From
this
dataset,
we
obtained
a
le
containing
the
title,
content,
creation
date,
and
the
truth
label
(real
or
f
ak
e)
for
each
article.
Using
this
data,
we
processed
all
inputs
according
to
the
w
orko
w
sho
wn
in
Figure
3.
Figure
3.
Data
cleaning
process:
steps
in
v
olv
ed
in
preparing
te
xt
data
for
analysis
The
methodology
be
gins
with
preprocessing
the
collected
data,
where
all
non-essential
elements
li
k
e
links,
emojis,
numbers,
and
special
characters
are
remo
v
ed.
The
remaining
te
xt
is
then
standardized
to
consist
only
of
the
alphabet.
F
ollo
wing
this,
we
tok
enize
the
te
xt
into
indi
vidual
w
ords
and
remo
v
e
common
stopw
ords
that
do
not
contrib
ute
meaningfully
to
f
ak
e
ne
ws
detection.
Adv
anced
techniques
lik
e
stemming
and
lemmati-
zation
are
applied
to
group
w
ords
by
their
root
forms.
F
or
e
xample,
”boating”
and
”boats”
are
reduced
to
their
base
form
”boat.
”
W
e
also
remo
v
e
numbers
and
links
(both
HTTP
and
HTTPS).
Ne
xt,
we
tok
enize
each
te
xt
by
con
v
erting
it
into
tok
ens.
F
or
e
xample,
the
sentence
”The
earth
is
round”
w
ould
be
tok
enized
into
”the,
”
”earth,
”
”is,
”
and
”round.
”
Ho
we
v
er
,
common
w
ords
lik
e
”the”
and
”is”
do
not
signicantly
contrib
ute
to
the
understanding
of
the
sentence’
s
meaning.
T
o
reduce
unnecessary
comple
xity
,
we
remo
v
e
stopw
ords—w
ords
lik
e
”the”
and
”is”—lea
ving
us
with
meaningful
tok
ens
such
as
”earth”
and
”round.
”
Compar
ative
analysis
of
mac
hine
learning
models
for
...
(Bahaa
Eddine
Elba
ghazaoui)
Evaluation Warning : The document was created with Spire.PDF for Python.
1956
❒
ISSN:
2252-8938
W
e
also
apply
stemming,
which
in
v
olv
es
reducing
w
ords
to
their
root
form.
F
or
instance,
a
search
for
”boat”
will
also
return
results
for
”boats,
”
”boating,
”
and
”boater
,
”
as
the
y
all
share
the
root
”boat.
”
In
contrast,
lemmatization
tak
es
into
account
the
broader
v
ocab
ulary
of
the
language,
con
v
erting
w
ords
lik
e
”w
as”
to
”be”
and
”mice”
to
”mouse.
”
After
stopw
ord
remo
v
al
and
stemming,
we
proceed
with
v
ectorizing
the
te
xt
using
term
frequenc
y-
in
v
erse
document
frequenc
y
(TF-IDF).
The
TF-IDF
method
assigns
a
weight
to
each
w
ord
in
the
document
based
on
ho
w
frequently
it
appears
within
the
document
term
frequenc
y
(TF)
and
ho
w
rare
the
w
ord
is
across
the
entire
corpus
in
v
erse
document
frequenc
y
(IDF).
This
approach
helps
emphasize
w
ords
that
are
important
to
a
particular
document
b
ut
not
common
across
man
y
documents.
The
result
is
a
numerical
representation
of
the
te
xt,
which
can
then
be
fed
into
machine
learni
ng
algorithms
for
classication.
F
or
e
xample,
in
the
sentence
”earth
is
round,
”
after
stopw
ord
remo
v
al
and
stemming,
the
im
p
or
tant
w
ord
”earth”
w
ould
ha
v
e
a
higher
TF-
IDF
score
if
it
appears
infrequently
in
the
dataset,
making
it
a
strong
feature
for
distinguishing
f
ak
e
ne
ws
from
real
ne
ws.
After
cleaning
the
dataset,
we
mo
v
ed
on
to
applying
machine
learning
models
to
the
processed
data.
The
input
to
our
models
consists
of
the
e
xtracted
w
ord
features
from
the
ne
ws
articles,
with
the
output
being
the
ne
ws
v
alidity
(f
ak
e
or
real).
W
e
implemented
v
arious
machine
learning
algorithms
using
Python
and
ran
them
on
the
same
machine,
producing
the
results
sho
wn
in
T
able
1.
T
able
1.
Model
comparison
for
f
ak
e
ne
ws
detection
Model
Accurac
y
(%)
Recall
(%)
F1-score
(%)
Duration
(s)
Linear
SVC
99.43
99.40
99.42
21.2639
Logistic
re
gression
98.89
98.85
98.87
38.6611
Multinomial
nai
v
e
Bayes
93.66
93.50
93.58
22.0979
Bernoulli
nai
v
e
Bayes
95.84
95.70
95.77
22.7570
Gradient
boost
99.61
99.60
99.60
159.8904
XGBoost
99.59
99.55
99.57
64.7849
Decision
tree
99.55
99.50
99.52
41.5180
K
nearest
neighbour
90.43
90.00
90.21
101.2192
Aggressi
v
e
classier
99.38
99.30
99.34
20.4679
The
results
sho
w
that
gradi
ent
boosting
achie
v
ed
the
highest
accurac
y
and
F1-score,
making
it
the
top-performing
model
in
terms
of
precision.
Ho
we
v
er
,
when
considering
e
x
ecution
time,
XGBoost
emer
ges
as
a
more
practical
choice,
pro
viding
nearly
the
s
ame
accurac
y
b
ut
with
signicantly
reduced
processing
time,
as
sho
wn
in
Figure
4
and
T
able
1.
This
trade-of
f
between
speed
and
accurac
y
is
critical
for
applications
requiring
real-time
analysis.
Additionally
,
the
Linear
SVC
model
,
though
slightly
less
accurate,
of
fers
the
f
astest
e
x
e-
cution,
which
can
be
benecial
in
resource-constrained
en
vironments.
The
main
chall
enges
encountered
were
optimizing
these
models
for
both
speed
and
accurac
y
,
particularly
when
w
orking
with
lar
ge
datasets.
Figure
4.
Comparison
of
machine
learning
models:
accurac
y
vs.
e
x
ecution
time
for
f
ak
e
ne
ws
detection
Int
J
Artif
Intell,
V
ol.
14,
No.
3,
June
2025:
1951–1959
Evaluation Warning : The document was created with Spire.PDF for Python.
Int
J
Artif
Intell
ISSN:
2252-8938
❒
1957
W
e
cannot
deniti
v
ely
say
that
gradient
boosting
is
optimal
model,
as
other
models
achie
v
ed
simi
lar
accurac
y
with
shorter
e
x
ecution
times.
It’
s
also
w
orth
noting
that
e
x
ecution
time
might
not
be
the
best
metric
for
model
comparison,
gi
v
en
dif
ferences
in
system
architectures.
Performance
can
be
impro
v
ed
by
using
GPUs
or
adding
more
cores
to
the
CPU,
potentially
reducing
e
x
ecution
time
and
further
optimizing
the
models.
8.
CONCLUSION
This
research
primarily
focused
on
addressing
the
four
fundamental
challenges
associated
with
big
data:
data
collection,
storage,
analysis,
and
prediction.
In
the
conte
xt
of
processing
te
xt
data,
we
implemented
v
arious
machine
learning
models
to
detect
f
ak
e
ne
ws
on
social
media
platforms.
Through
a
comparati
v
e
anal-
ysis
of
se
v
eral
models,
we
determined
which
model
best
met
our
requirements
based
on
accurac
y
and
perfor
-
mance.
Our
ndings
indicate
that
the
gradient
boosting
model
is
the
most
ef
f
ecti
v
e,
achie
vi
ng
an
accurac
y
of
99.61%
in
predicting
f
ak
e
ne
ws.
Additionally
,
the
XGBoost
model
also
demonstrated
high
performance,
with
an
accurac
y
of
99.59%.
XGBoost
pro
vided
a
strong
balance
between
accurac
y
and
e
x
ecution
time,
making
it
an
e
xcellent
choice
for
applications
where
both
precision
and
speed
are
critical.
The
consistent
performance
of
both
models
highlights
their
rob
ustness
in
handling
comple
x
data
and
deli
v
ering
reliable
predictions.
This
study
is
limited
by
its
focus
on
te
xt-based
f
ak
e
ne
ws
detection,
which
does
not
account
for
the
multimedia
elements
often
in
v
olv
ed
in
misinformation,
such
as
images
or
videos.
Additionally
,
the
dataset
used,
while
comprehen-
si
v
e,
is
constrained
to
specic
time
periods
and
domains.
Future
research
should
e
xplore
inte
grating
image
and
video
analysis
into
f
ak
e
ne
ws
detection
systems
to
impro
v
e
accurac
y
further
.
Moreo
v
er
,
e
xpanding
the
scope
to
real-time
detection
could
pro
vide
more
dynamic
and
responsi
v
e
solutions
for
combating
misinformation
on
social
media
platforms.
FUNDING
INFORMA
TION
The
authors
declare
that
no
specic
funding
w
as
recei
v
ed
for
this
research.
A
UTHOR
CONTRIB
UTIONS
ST
A
TEMENT
This
journal
uses
the
Contrib
utor
Roles
T
axonomy
(CRediT)
to
recognize
indi
vidual
author
contrib
u-
tions,
reduce
authorship
disputes,
and
f
acilitate
collaboration.
Name
of
A
uthor
C
M
So
V
a
F
o
I
R
D
O
E
V
i
Su
P
Fu
Bahaa
Eddine
Elbaghazaoui
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Mohamed
Amnai
✓
✓
✓
✓
✓
✓
✓
Y
oussef
F
akhri
✓
✓
✓
✓
✓
✓
✓
✓
Ali
Choukri
✓
✓
✓
✓
✓
✓
✓
Noreddine
Gherabi
✓
✓
✓
✓
✓
✓
✓
✓
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
kno
wn
competing
nancial
interests
or
personal
relati
onships
that
could
ha
v
e
appeared
to
inuence
the
w
ork
reported
in
this
paper
.
D
A
T
A
A
V
AILABILITY
The
data
used
in
this
study
is
the
ISO
T
f
ak
e
ne
ws
dataset,
created
by
the
Information
Security
and
Object
T
echnology
(ISO
T)
research
group
at
the
Uni
v
ersity
of
V
ictoria.
This
dataset
is
publicly
a
v
ailable
and
widely
used
for
f
ak
e
ne
ws
detection
tasks.
It
consists
of
tw
o
cate
gories:
real
ne
ws
and
f
ak
e
ne
ws.
The
dataset
Compar
ative
analysis
of
mac
hine
learning
models
for
...
(Bahaa
Eddine
Elba
ghazaoui)
Evaluation Warning : The document was created with Spire.PDF for Python.
1958
❒
ISSN:
2252-8938
can
be
accessed
at:
https://www
.kaggle.com/datasets/emine
yetm/f
ak
e-ne
ws-detection-datasets.
REFERENCES
[1]
M
.
Cai,
H.
Luo,
X.
Meng,
Y
.
Cui,
and
W
.
W
ang,
“Inuence
of
information
attrib
utes
on
information
dissemination
in
public
health
emer
gencies,
”
Humanities
and
Social
Sciences
Communications
,
v
ol.
9,
no.
1,
pp.
1-22,
2022,
doi:
10.1057/s41599-022-01078-3.
[2]
E.
C.
T
andoc,
“F
ak
e
ne
ws,
”
in
The
Routledg
e
Companion
to
Media
Disinformat
ion
and
P
opulism
,
London,
United
Kingdom:
Routledge,
2021,
pp.
110-117.
[3]
S
.
V
.
D.
Linden,
“Misinformation:
susceptibility
,
spread,
and
interv
entions
to
immunize
the
public,
”
Natur
e
Medicine
,
v
ol.
28,
no.
3,
pp.
460-467,
2022,
doi:
10.1038/s41591-022-01713-6.
[4]
J
.
Geor
ge,
N.
Gerhart,
and
R.
T
orres,
“Unco
v
ering
the
truth
about
f
ak
e
ne
ws:
A
research
model
grounded
in
multi-disciplinary
literature,
”
F
ak
e
Ne
ws
on
the
Internet
,
pp.
175-202,
2023.
[5]
R
.
Rodr
´
ıguez-Ferr
´
andiz,
“
An
o
v
ervie
w
of
the
f
ak
e
ne
ws
phenomenon:
From
untruth-dri
v
en
to
post-truth-dri
v
en
approaches,
”
Media
and
Communication
,
v
ol.
11,
no.
2,
pp.
15-29,
2023,
doi:
10.17645/mac.v11i2.6209.
[6]
P
.
N.
Petratos,
“Misinformati
on,
disinformation,
and
f
ak
e
ne
ws:
Cyber
risks
to
b
usiness,
”
Business
Horizons
,
v
ol.
64,
no.
6,
pp.
763-774,
2021,
doi:
10.1016/j.b
ushor
.2021.07.010.
[7]
E.
A
¨
ımeur
,
S.
Amri,
and
G.
Brassard,
“F
ak
e
ne
ws,
disinformation
and
misinformation
in
social
media:
a
re
vie
w
,
”
Social
Network
Analysis
and
Mining
,
v
ol.
13,
no.
1,
2023,
doi:
10.1007/s13278-023-00951-3.
[8]
X.
Zhang
and
A.
A.
Ghorbani,
“
An
o
v
ervie
w
of
online
f
ak
e
ne
ws:
Characterization,
detection,
and
discussion,
”
Information
Pr
o-
cessing
&
Mana
g
ement
,
v
ol.
57,
no.
2,
2020,
doi:
10.1016/j.ipm.2019.102025.
[9]
D.
Ev
ans,
S.
Bratton,
and
J
.
McK
ee,
Social
Media
Mark
eting:
the
ne
xt
g
ener
ation
of
b
usiness
enga
g
ement
,
Hobok
en,
United
States:
A
G
Printing
&
Publishing,
2021.
[10]
G.
D.
Domenico,
J.
Sit,
A.
Ishizaka,
and
D.
Nunan,
“F
ak
e
ne
ws,
social
media
and
mark
eting:
A
systematic
re
vie
w
,
”
J
ournal
of
Business
Resear
c
h
,
v
ol.
124,
pp.
329-341,
2021,
doi:
10.1016/j.jb
usres.2020.11.037.
[11]
J.
P
.
Baptista
and
A.
Gradim,
“Understanding
f
ak
e
ne
ws
consumption:
A
re
vie
w
,
”
Social
Sciences
,
v
ol.
9,
no.
10,
2020,
doi:
10.3390/socsci9100185.
[12]
G.
Penn
ycook
and
D.
G.
Rand,
“The
psychology
of
f
ak
e
ne
ws,
”
T
r
ends
in
Co
gnitive
Sciences
,
v
ol.
25,
no.
5,
pp.
388-402,
2021,
doi:
10.1016/j.tics.2021.02.007.
[13]
R.
Armitage
and
C.
V
accari,
“Misinformation
and
disinformation,
”
in
The
Routledg
e
Companion
to
Media
Disinformation
and
P
opulism
,
London,
United
Kingdom:
Routledge,
2021.
[14]
I.
P
.
Idok
o,
O.
M.
Ijig
a,
L.
A.
En
yejo,
S.
I.
Ugbane,
O.
Ak
oh,
and
M.
O.
Ode
yemi,
“Exploring
the
pot
ential
of
Elon
musk’
s
proposed
quantum
AI:
A
comprehensi
v
e
analysis
and
im
plications,
”
Global
J
ournal
of
Engineering
and
T
ec
hnolo
gy
Advances
,
v
ol.
18,
no.
3,
pp.
48-65,
2024,
doi:
10.30574/gjeta.2024.18.3.0462.
[15]
O.
D.
Apuk
e
and
B.
Omar
,
“F
ak
e
ne
ws
and
CO
VID-19:
Modelli
ng
the
predictors
of
f
ak
e
ne
ws
sharing
among
social
media
users,
”
T
elematics
and
Informatics
,
v
ol.
56,
2021,
doi:
10.1016/j.tele.2020.101475.
[16]
Y
.
M.
Rocha,
G.
A.
de
Moura,
G.
A.
Desid
´
erio,
C.
H.
de
Oli
v
eira,
F
.
D.
Lourenc
¸
o,
and
L.
D.
F
.
Nicolete,
“The
impact
of
f
ak
e
ne
ws
on
social
media
and
its
i
nuence
on
health
during
the
CO
VID-19
pandemic:
a
systematic
re
vie
w
,
”
J
ournal
of
Public
Health
,
v
ol.
31,
pp.
1007–1016,
2021,
doi:
10.1007/s10389-021-01565-3.
[17]
A.
Mishra
and
H.
Sadia,
“
A
comprehensi
v
e
analysis
of
f
ak
e
ne
ws
detection
models:
A
systematic
literature
re
vie
w
and
current
challenges,
”
Engineering
Pr
oceedings
,
v
ol.
59,
no.
1,
2023,
doi:
10.3390/engproc2023059028.
[18]
E.
Min
et
al
.,
“Di
vide-and-conquer:
Post-user
interaction
netw
ork
for
f
ak
e
ne
ws
detection
on
social
media,
”
in
Pr
oceedings
of
the
A
CM
W
eb
Confer
ence
,
2022,
pp.
1148-1158,
doi:
10.1145/3485447.3512163.
[19]
P
.
Meel
and
D.
K.
V
ishw
akarma,
“F
ak
e
ne
ws,
rumor
,
information
pollution
in
social
media
and
web:
A
contempora
ry
surv
e
y
of
state-of-the-arts,
challenges
and
opportunities,
”
Expert
Systems
with
Applications
,
v
ol.
153,
2020,
doi:
10.1016/j.esw
a.2020.112986.
[20]
C.
B
usioc,
S.
Rus
eti,
and
M.
Dascal
u,
“
A
literature
re
vie
w
of
NLP
approaches
to
f
ak
e
ne
ws
detection
and
their
applicability
to
Romanian
language
ne
ws
analysis,
”
Re
vista
T
r
ansilvania
,
v
ol.
10,
pp.
65-71,
2020,
doi:
10.51391/trv
a.2020.10.07.
[21]
K.
Bryano
v
and
V
.
Vziatyshe
v
a,
“Determinants
of
indi
viduals’
belief
in
f
ak
e
ne
ws
:
A
scoping
re
vie
w
,
”
PLOS
ONE
,
v
ol.
16,
no.
6,
2021,
doi:
10.1371/journal.pone.0253717.
[22]
B.
E.
Elbaghazaoui,
M.
Amnai,
and
Y
.
F
akhri,
“Predicting
user
beha
vior
using
data
proling
and
hidden
Mark
o
v
model,
”
Interna-
tional
J
ournal
of
Electrical
&
Computer
Engineering
,
v
ol.
13,
no.
5,
pp.
4819-4827,
2023,
doi:
10.11591/ijece.v13i5.pp5444-5453
[23]
B.
E.
Elbaghazaoui,
M.
Amnai,
Y
.
F
akhri,
A.
Choukri
and
N.
Gherabi,
“Ev
ent
proling
in
social
media:
A
surv
e
y
with
an
appli-
cation,
”
in
2024
4th
International
Confer
ence
on
Inno
vative
Resear
c
h
in
Applied
Science
,
Engineeri
ng
and
T
ec
hnolo
gy
(IRASET)
,
FEZ,
Morocco,
2024,
pp.
1-5,
doi:
10.1109/IRASET60544.2024.10548943.
[24]
B.
E.
Elbaghzaoui,
M.
Amnai,
and
Y
.
F
akhri,
“Predicting
the
ne
xt
w
ord
using
the
Mark
o
v
chain
model
according
to
proling
personality
,
”
The
J
ournal
of
Super
computing
,
v
ol.
79,
no.
11,
pp.
12126-12141,
2023,
doi:
10.1007/s11227-023-04964-7.
[25]
A.
A.
A.
Ahmed,
A.
Aljabouh,
P
.
K.
Donepudi,
and
M.
S.
Choi,
“Detecting
f
ak
e
ne
ws
using
machine
learning:
A
systematic
literature
re
vie
w
,
”
arXiv-Computer
Science
,
pp.
1-10,
2021.
[26]
R.
K.
Kaliyar
,
A.
Gosw
am
i,
and
P
.
Narang,
“DeepF
akE:
Impro
ving
f
ak
e
ne
ws
detec
tion
using
tensor
decomposition-based
deep
neural
netw
ork,
”
The
J
ournal
of
Super
computing
,
v
ol.
77,
no.
2,
pp.
1015-1037,
2021,
doi:
10.1007/s11227-020-03294-y
.
[27]
S.
Rathje,
J.
Roozenbeek,
J.
J.
V
.
Ba
v
el,
and
S.
V
.
D.
Linden,
“
Accurac
y
and
social
moti
v
ations
shape
judgements
of
(mis)information,
”
Natur
e
Human
Behaviour
,
v
ol.
7,
no.
6,
pp.
892-903,
2023,
doi:
10.1038/s41562-023-01594-9.
[28]
M.
T
.
Sadiq
and
S.
K.
Mathe
w
,
“The
disaster
of
misinformation:
A
re
vie
w
of
research
in
social
media,
”
International
J
ournal
of
Data
Science
and
Analytics
,
v
ol.
13,
no.
4,
pp.
271-285,
2022,
doi:
10.1007/s41060-022-00321-3.
[29]
H.
Ahmed,
I.
T
raore,
and
S.
Saad,
“Detecting
opinion
spams
and
f
ak
e
ne
ws
using
te
xt
classication,
”
Security
and
Privacy
,
v
ol.
1,
no.
1,
2017,
doi:
10.1002/sp
y2.9.
[30]
H.
Ahmed,
I.
T
raore,
and
S.
Saad,
“Detection
of
online
f
ak
e
ne
ws
using
N-Gram
analysis
and
machine
learning
techniques,
”
in
Intellig
ent,
Secur
e
,
and
Dependable
Systems
in
Distrib
uted
and
Cloud
En
vir
onments
,
Springer
,
Cham,
2017,
pp.
127-138,
doi:
10.1007/978-3-319-69155-8
9.
Int
J
Artif
Intell,
V
ol.
14,
No.
3,
June
2025:
1951–1959
Evaluation Warning : The document was created with Spire.PDF for Python.
Int
J
Artif
Intell
ISSN:
2252-8938
❒
1959
BIOGRAPHIES
OF
A
UTHORS
Bahaa
Eddine
Elbaghazaoui
be
g
an
his
academic
journe
y
in
mathematical
sciences,
earning
his
baccalaureate
in
this
eld.
After
completing
preparatory
classes
in
2013,
he
adv
anced
to
the
National
School
of
Applied
Sciences
of
Khouribg
a,
where
he
pursued
computer
engineer
-
ing,
ulti
mately
earning
an
engineering
diploma
in
softw
are
engineering
in
2019.
In
December
2023,
he
obtained
his
Ph.D.
in
computer
sciences
&
articial
intelligence.
W
ith
o
v
er
5
years
of
e
xperience
as
a
softw
are
engineer
and
full-stack
de
v
eloper
,
he
continues
to
contrib
ute
to
the
eld
through
his
w
ork
and
research.
He
is
no
w
an
Assistant
Professor
at
ENSA
Beni
Mellal.
He
can
be
contacted
at
email:
elbaghazaoui.bahaa@gmail.com,
bahaaeddine.elba
ghazaoui@uit.ac.ma,
or
b
.elbaghazaoui@usms.ac.ma.
Mohamed
Amnai
completed
his
under
graduate
studies
in
computers,
elect
ronics,
electri-
cal,
and
automation
at
Molay
Ismail
Uni
v
ersity
in
Errachidia
in
2000.
He
later
obtained
his
master’
s
de
gree
in
the
same
eld
from
Ibn
T
of
ail
Uni
v
ersit
y
in
K
enitra
in
2007.
In
2011,
he
earned
a
Ph.D.
in
computer
science
and
telecommunications
from
Ibn
T
of
ail
Uni
v
ersity
.
Since
2014,
he
has
been
serving
as
an
Assistant
Professor
at
the
National
School
of
Applied
Sciences
Khouribg
a
of
Settat
Uni
v
ersity
.
In
2018,
he
w
as
appointed
as
an
Associate
Professor
in
the
Department
of
Computer
Science
and
Mathe
matics
at
the
F
aculty
of
Sciences
in
K
enitra,
Ibn
T
of
ail
Uni
v
ersity
.
He
is
also
a
member
of
the
Netw
orks
and
T
elecommunications
T
eam
at
the
K
enitra
F
aculty
of
Science
and
a
researcher
in
the
LaRIT
Laboratory
.
He
can
be
contacted
at
email:
mohamed.amnai@uit.ac.ma.
Y
oussef
F
akhri
obtained
his
Bachelor
of
Science
de
gre
e
in
electronic
ph
ysics
from
the
Uni
v
ersity
Mohammed
V’
s
F
aculty
of
Sciences
in
Rabat
in
2001.
He
then
pursued
a
master’
s
de
gree
in
computer
and
telecommunication,
where
he
completed
a
project
with
the
Moroccan
ICI
Com-
pan
y
in
2003.
In
2007,
he
recei
v
ed
his
Ph.D.
from
the
Uni
v
ersity
Mohammed
V
-Agdal
in
Rabat,
Morocco,
in
collaboration
with
the
Polytechnic
Uni
v
ersity
of
Catalonia
in
Spain.
F
ollo
wing
his
aca-
demic
achie
v
ements,
he
joined
Ibn
T
of
ail
Uni
v
ersity’
s
F
aculty
of
Sciences
in
K
enitra
as
an
Associate
Professor
in
2009,
where
he
currently
teaches
in
the
Department
of
Compute
r
Science
and
Mathemat-
ics.
He
also
serv
es
as
an
associate
researcher
at
the
Rabat
F
aculty
of
Sciences
and
is
the
Laboratory
Head
at
LaRIT
.
He
can
be
contacted
at
email:
f
akhri@uit.ac.ma.
Ali
Choukri
is
an
Assistant
Professor
at
the
National
Academy
of
Applied
Sciences.
He
recei
v
ed
his
mast
er’
s
de
gree
in
computer
science
and
telecommunic
ations
from
the
Uni
v
ersity
of
Ibn
T
of
ail,
K
enitra,
Morocco,
in
2008,
and
in
1992,
he
earned
a
de
gree
from
ENSET
(Higher
Normal
School
of
T
echnical
T
eaching).
He
hol
ds
a
Ph.D.
from
the
School
of
Computer
Science
and
Systems
Analysis
(ENSIAS).
He
is
part
of
the
MIS
team
in
the
SIME
Laboratory
,
where
his
research
focuses
on
mobile
intelligent
ad
hoc
communication
systems
and
wireles
s
sensor
netw
orks.
His
research
interests
span
a
wide
range
of
topics,
including
ubiquitous
computing,
the
internet
of
things,
delay/f
ault-tolerant
netw
orks,
wireless
netw
orks,
QoS
routing,
mathematical
modeling
and
performance
analysis
of
netw
orks,
control
and
decision
theory
,
g
ame
theory
,
trust
and
reputation
management,
distrib
uted
algorithms,
meta-heuristics
and
optimization,
and
genetic
algorithms.
He
can
be
contacted
at
email:
choukriali@gmail.com.
Nor
eddine
Gherabi
is
a
Professor
of
computer
science
with
e
xtensi
v
e
e
xperience
i
n
both
academia
and
industry
.
He
holds
a
doctorate
in
computer
science
and
has
been
teaching
at
Mohamed
Ben
Abdellah
Uni
v
ersity
since
2013,
and
at
Sultan
Moulay
Sli
mane
Uni
v
ersity
in
Morocco
as
a
research
professor
since
2015.
A
member
of
the
International
Association
of
Engineers,
he
has
made
signicant
contrib
utions
to
information
systems,
focusing
on
big
data,
semantic
web,
pattern
recognition,
and
intelligent
systems.
He
has
numerous
publications
in
computer
science,
including
book
chapters,
journal
artic
les,
and
conference
papers,
and
has
edited
se
v
eral
books.
He
has
also
chaired
and
con
v
ened
o
v
er
52
conferences
and
w
orkshops.
His
recent
books
published
by
Springer
include
intelligent
systems
in
big
data,
semantic
web,
and
machine
learning,
adv
ances
in
information,
communication,
c
ybersecurity
,
and
information
technology
and
communication
systems.
He
can
be
contacted
at
email:
gherabi@gmail.com
or
n.gherabi@usms.ma.
Compar
ative
analysis
of
mac
hine
learning
models
for
...
(Bahaa
Eddine
Elba
ghazaoui)
Evaluation Warning : The document was created with Spire.PDF for Python.