IAES
Inter
national
J
our
nal
of
Articial
Intelligence
(IJ-AI)
V
ol.
15,
No.
4,
August
2026,
pp.
3164
∼
3175
ISSN:
2252-8938,
DOI:
10.11591/ijai.v15.i4.pp3164-3175
❒
3164
Exploring
cogniti
v
e
patter
ns
in
childr
en
with
autism
spectrum
disorder
using
corr
elation
and
cluster
analysis
Hana
Bezzih
1
,
Muna
Darweesh
2
,
Amjad
Gawanmeh
2
1
Laboratoire
Applied
Research
in
Psychology
and
Education,
Uni
v
ersity
Mohammed
Lamine
Dabaghine,
Setif,
Algeria
2
Colle
ge
of
Engineering
and
Information
T
echnology
,
Uni
v
ersity
of
Dubai,
Dubai,
United
Arab
Emirates
Article
Inf
o
Article
history:
Recei
v
ed
No
v
15,
2024
Re
vised
May
14,
2026
Accepted
Jul
9,
2026
K
eyw
ords:
Autism
spectrum
disorder
Autism
spectrum
disorder
assessment
Cogniti
v
e
de
v
elopment
Correlation
and
cluster
analysis
V
isual-spatial
skills
ABSTRA
CT
This
paper
in
v
estig
ates
the
relationships
between
age
and
v
e
cogniti
v
e
abilities
in
children
diagnosed
with
autism
spec
trum
disorder
(ASD).
Data
f
rom
210
children
aged
6
to
12
years
who
completed
measures
of
visual
motor
precision
(VM),
f
acial
memory
(FM),
spatial
vision
(SV),
dra
wing
memory
(DM),
and
orientation
(OR)
were
analyzed.
Data
preparation,
descripti
v
e
statistical
analysis,
correlation
analysis,
cluster
analysis,
and
f
actor
analysis
were
conducted.
The
results
sho
wed
that
age
w
as
not
signicantly
associated
with
the
cogniti
v
e
measures.
The
cle
arest
association
w
as
a
moderate
positi
v
e
correlation
between
VM
and
FM
(0.34).
Cluster
analysis
ident
ied
three
groups.
Ho
we
v
er
,
the
lo
w
silhouette
score
(0.1368)
indicated
weak
separation
and
substantial
o
v
erlap
between
cogniti
v
e
proles.
F
actor
analysis
suggested
three
latent
dimensions:
f
actor
1
w
as
strongly
associated
with
VM
and
FM;
f
actor
2
w
as
ne
g
ati
v
ely
related
to
DM
and
moderately
pos
iti
v
ely
related
to
OR;
and
f
actor
3
w
as
primarily
dri
v
en
by
SV
.
These
ndings
suggest
that
cogniti
v
e
abilities
in
children
with
ASD
operate
relati
v
ely
independently
,
with
some
shared
mechanisms
between
certain
skills.
It
can
be
concluded
that
indi
vidualized
approaches
in
assessment
and
interv
ention
strate
gies
are
w
arranted.
This
is
an
open
access
article
under
the
CC
BY
-SA
license
.
Corresponding
A
uthor:
Amjad
Ga
w
anmeh
Colle
ge
of
Engineering
and
Information
T
echnology
,
Uni
v
ersity
of
Dubai
Dubai,
United
Arab
Emirates
Email:
amjad.g
a
w
anmeh@ieee.or
g
1.
INTR
ODUCTION
Cogniti
v
e
v
ariability
is
a
core
feature
of
autism
spectrum
disorder
(ASD),
a
neuro-de
v
elopmental
condition
characterized
by
challenges
in
social
interaction,
communication,
and
repetiti
v
e
beha
viors.
Children
with
ASD
often
e
xhibit
wide-ranging
dif
ferences
in
cogniti
v
e
abilities,
impacting
learning,
and
daily
functioning.
K
e
y
cogniti
v
e
abilities,
including
visual
motor
precision
(VM),
f
acial
memory
(FM)
[1],
spatial
vision
(SV)
[2],
dra
wing
memory
(DM)
[3],
and
orientation
(OR)
[4],
play
crucial
roles
in
a
child’
s
capacity
to
eng
age
with
and
comprehend
their
surroundings.
Understanding
the
relationships
between
these
cogniti
v
e
skills
could
identify
de
v
elopmental
needs
of
children
with
ASD
and
f
acilitate
the
design
of
tar
geted
interv
ention
and
educational
strate
gies
[5].
VM
refers
to
the
coordination
of
visual
perception
and
motor
action,
for
e
xample,
during
writing,
tracing,
or
dra
wing.
Dif
culties
in
VM
may
interfere
with
tasks
that
require
hand-e
ye
coordination
and
ne
motor
control.
FM
refers
to
the
ability
to
recognize
and
remember
f
aces
and
is
closely
link
ed
to
social
interaction.
Dif
culties
in
FM
may
contrib
ute
to
challenges
in
recognizing
people
and
interpreting
social
cues.
J
ournal
homepage:
http://ijai.iaescor
e
.com
Evaluation Warning : The document was created with Spire.PDF for Python.
Int
J
Artif
Intell
ISSN:
2252-8938
❒
3165
SV
refers
to
the
ability
to
percei
v
e
spatial
relations
between
objects
and
the
self,
which
is
needed
for
na
vig
ation
and
understanding
spatial
en
vironments.
DM
measures
the
ability
to
recall
and
reproduce
visual
forms
or
patterns,
reecting
visual
memory
and
visual
processing.
OR
refers
to
spatial
a
w
areness
and
the
ability
to
locate
oneself
or
objects
accurately
in
space.
T
ogether
,
these
abilities
pro
vide
a
useful
prole
of
cogniti
v
e
functioning
in
children
with
ASD.
Pre
vious
research
suggests
that
cogniti
v
e
abilities
in
ASD
are
heterogeneous,
with
children
sho
wing
dif
ferent
patterns
of
strengths
and
weaknesses
across
domains
[6],
[7].
Ho
we
v
er
,
the
relationships
between
these
abilities
remain
unclear
.
It
is
not
yet
clear
whether
stronger
performance
in
one
cogniti
v
e
domain
is
associated
with
st
ronger
performance
in
another
,
or
whether
these
domains
de
v
elop
lar
gely
independently
.
Therefore,
this
study
e
xamined
the
correlations
between
age,
VM,
FM,
SV
,
DM,
and
OR
in
a
sample
of
210
children
with
ASD
aged
6
to
12
years.
This
age
range
w
as
selected
because
it
represents
an
important
de
v
elopmental
period
during
which
cogniti
v
e
abilities
are
acti
v
ely
shaping
learning,
communication,
and
adapti
v
e
functioning.
T
o
e
xamine
the
relationships
between
the
cogniti
v
e
measures,
this
st
u
dy
e
xtends
the
pre
vious
results
[8]
and
use
Pearson
and
Spearman
correlation
matrices
to
assess
both
linear
and
rank-based
relationships
between
age
and
the
v
e
cogniti
v
e
v
ariables.
This
approach
allo
wed
us
to
determine
whether
specic
abilities
tended
to
co-v
ary
,
and
whether
age
w
as
as
sociated
with
cogniti
v
e
performance
within
the
sample.
Identifying
such
associations
is
important
because
related
abilities
may
share
underlying
mechanisms
or
may
impro
v
e
together
during
interv
ention.
This
study
then
used
K-means
cluster
analysis
to
e
xamine
whether
children
with
ASD
could
be
grouped
according
to
their
cogniti
v
e
performance
across
age,
VM,
FM,
SV
,
DM,
and
OR.
This
method
partitions
cases
into
groups
according
to
similarity
ac
ross
the
selected
v
ariables
and
can
re
v
eal
possible
cogniti
v
e
proles.
The
dataset
sho
wed
v
ariation
across
the
v
e
cogniti
v
e
dom
ains.
Clustering
quality
w
as
e
v
aluated
using
the
silhouette
score,
which
reects
ho
w
well
each
child
ts
within
the
assigned
cluster
compared
with
other
clusters.
Cluster
distrib
ution
w
as
visual
ized
using
principal
component
analysis
(PCA)
to
reduce
dimensionality
.
Additionally
,
f
actor
analysis
w
as
conducted
to
e
xplore
relationships
between
cogniti
v
e
v
ariables,
with
the
ai
m
of
pro
viding
a
comprehensi
v
e
understanding
of
cogniti
v
e
abilities
and
their
correlations
in
children
with
ASD.
The
ndings
could
inform
the
de
v
elopment
of
tar
geted
interv
entions
that
address
specic
cogniti
v
e
strengths
and
weaknesses,
ultimately
enhancing
educational
outcomes
and
capabilities
for
children
on
the
autism
spectrum.
2.
RELA
TED
W
ORK
There
has
been
increasing
interest
in
studying
se
v
eral
aspects
of
ASD
using
art
icial
intel
ligence
(AI)
methods.
V
arious
AI
applications
and
algorithms
ha
v
e
been
utilized
to
address
challenges
f
aced
by
indi
viduals
with
ASD,
including
early
diagnosis,
assessment,
personalized
interv
entions,
support,
and
educational
methods.
A
comprehensi
v
e
analysis
by
Salamanca
et
al.
[9]
e
xamined
e
x
ecuti
v
e
function
(EF)
proles
in
children
with
ASD,
re
v
ealing
signicant
decits
across
multiple
EF
domains.
The
children
demonstrated
dif
culties
in
cogniti
v
e
e
xibility
and
planning.
These
ndings
highlight
the
need
for
tar
geted
interv
entions
aimed
at
impro
ving
EF
skills
in
children
with
ASD.
In
the
area
of
visual-spatial
processing,
there
is
e
vidence
that
children
with
ASD
may
utilize
alte
rnati
v
e
strate
gies
for
processing
visual-spatial
information.
Moug
a
et
al.
[10]
used
adv
anced
e
ye-tracking
technology
to
g
ain
insights
into
ho
w
children
with
ASD
percei
v
e
and
interpret
spatial
information.
Their
study
unco
v
ered
atypical
g
aze
patterns
during
spatial
tasks,
suggesting
t
hat
these
children
may
dra
w
on
dif
ferent
cogniti
v
e
resources
when
processing
spatial
relationships
compared
with
typically
de
v
eloping
peers.
This
research
has
important
implications
for
t
he
de
v
elopment
of
educational
materials
tailored
to
le
v
erage
the
visual-spatial
processing
abilities
observ
ed
in
children
with
ASD.
Social
cognition,
a
k
e
y
area
of
dif
culty
in
ASD,
w
as
focus
of
a
longitudinal
study
[11].
Ov
er
a
v
e-year
period,
the
y
track
ed
the
de
v
elopment
of
social
cogniti
v
e
skills
in
children
with
ASD.
While
the
y
observ
ed
impro
v
ements
in
these
skills
o
v
er
time,
the
rate
of
progress
w
as
slo
wer
compared
to
typically
de
v
eloping
peers,
with
theory
of
mind
abilities
sho
wing
the
most
signicant
lag.
These
ndings
are
consistent
with
pre
vious
research
reporting
social
cogniti
v
e
dif
ferences
in
autism
[12],
and
emphasize
the
importance
of
early
and
sustained
interv
entions
tar
geting
social
cogniti
v
e
skills
in
ASD.
Se
v
eral
studies
ha
v
e
e
xplored
v
arious
aspects
of
cogniti
v
e
and
beha
vioral
proles
in
indi
viduals
with
ASD.
Vries
et
al.
[13]
conducted
a
randomized
controlled
trial
to
e
v
al
uate
the
ef
cac
y
of
cogniti
v
e
e
xibility
Exploring
co
gnitive
patterns
in
c
hildr
en
with
autism
spectrum
disor
der
using
corr
elation
and
...
(Hana
Bezzih)
Evaluation Warning : The document was created with Spire.PDF for Python.
3166
❒
ISSN:
2252-8938
training
in
indi
viduals
with
ASD.
The
study
aimed
to
impro
v
e
this
critical
EF
in
ASD
participants,
and
the
research
assessed
the
impact
of
tar
geted
interv
entions
on
cogniti
v
e
e
xibility
,
a
skill
often
found
to
be
challenging
for
indi
viduals
with
ASD.
Gabrielsen
et
al.
[14]
utilized
cluster
analysis
to
in
v
estig
ate
visual
cogniti
v
e
styles
in
autism,
potentially
identifying
subgroups
according
t
o
visual
processing
characteristics,
brain
imagi
ng
techniques
were
emplo
yed
to
in
v
estig
ate
neural
mechanisms
associat
ed
with
language
dif
culties
in
this
subgroup.
The
study
demonstrated
that
language
processing
in
these
children
dif
fers
from
typical
neural
processing
patterns.
Brigido
et
al.
[15]
also
utilized
cluster
analysis,
focusing
on
beha
vioral
proles
in
ASD
in
order
to
identify
distinct
beha
vioral
subgroups
within
the
autism
spectrum.
Rosello
et
al.
[16]
conducted
a
longitudinal
study
e
xamining
e
x
ecuti
v
e
and
socio-adapti
v
e
beha
viors
in
adolescents
with
ASD
without
intellectual
disability
.
T
racking
the
de
v
elopment
of
these
skills
o
v
er
time
and
identifying
distinct
subgroups
within
this
ASD
population,
the
study
sho
ws
ho
w
EF
and
social
adaptation
skills
e
v
olv
e
during
adolescence
in
dif
ferent
ASD
subgroups,
potentially
informing
more
tailored
interv
entions
and
support
strate
gies.
T
able
1
sho
ws
a
summary
of
research
studies
on
ASD.
T
able
1.
Summary
of
research
studies
on
ASD
Ref.
Study
focus
Methods
and
tools
Benets
Limitations
[17]
Educational
support
for
childre
n
with
ASD
Cogniti
v
e
computing
machine
learning
(ML)
Personalized
learning
and
adapti
v
e
assistance
Limited
scalability
,
bias,
and
de
v
elopment
intensi
v
e
[12]
Perception
enhancement
ML
and
wearable
sensors
robotic
systems
Impro
v
ed
perception
and
social
interaction
Cost,
discomfort,
and
limited
accessibility
[18]
Early
diagnosis
and
screening
ML
data
mining
Accurate
as
sessment
and
early
interv
ention
Lack
of
specicity
[19]
Healthcare
applications
AI
and
data
analysis
Enhanced
healthcare
management
and
personalized
treatment
Data
constraints
and
potential
for
pri
v
ac
y
concerns
[20]
Screening
and
early
detection
AI
algorithms
and
data
analysis
Ef
cient
screening
and
early
interv
ention
Limited
scope
and
v
alidation
needed
[21]
Classication
and
detection
Deep
learning
Accurate
and
early
identication
Requires
lar
ge
datasets
[22]
ASD
diagnosis
ML
correlation
lters
Diagnostic
accurac
y
and
personalized
treatment
Limited
and
Feature-dependent
[23]
F
ace
response
classication
ML
neuroimaging
Identication
of
neural
abnormalities
and
early
diagnosis
Expensi
v
e,
time-consuming,
and
limited
accessibility
[24]
Functional
connections
ML
optimization
and
neuroimaging
Brain
netw
ork
abnormalities
and
personalized
treatment
Expensi
v
e,
time-consuming,
and
visual-specic
[15]
Beha
vioral
proles
in
ASD
Cluster
analysis
distinct
subgroups
within
ASD
Cannot
capture
all
ASD
dimensions
[25]
V
isual
cogniti
v
e
s
tyles
in
ASD
Cluster
analysis
V
isual
processing
patterns
Limited
to
visual
cognition
[26]
Gesture
skills
in
ASD
Cluster
analysis
Gesture
skil
l-based
subgroups
F
ocuses
on
Chinese
autistic
children
[16]
Ex
ecuti
v
e
beha
viors
Statistical
analysis
Ex
ecuti
v
e,
socio-adapti
v
e
beha
viors
Children
without
intellectual
disability
[27]
Diagnosis
and
assessment
Fuzzy
cogniti
v
e
maps
modeling
softw
are
Impro
v
ed
diagnostic
accurac
y
Limited
generalization
and
potential
for
o
v
ertting
[10]
V
isual-spatial
process
Eye-tracking
and
AI
Processing
insi
ghts
Eye-tracking
only
[11]
Social
cognition
in
ASD
Statistical
analysis
Long-term
insights
Non-AI
approach
and
limited
scope
[28]
Electroencephalogram
(EEG)
ASD
prediction
Deep
learning
and
con
v
olutional
neural
netw
ork
(CNN)
Early
detection
EEG-dependent
[29]
Emotion
recognition
Computer
vision
and
ML
Impro
v
e
social
int
eraction
F
acial
e
xpression
focus
[30]
ASD
t
raits
classication
Natural
language
processing
(NLP)
and
transformer
models
Language-based
insights
T
e
xt
data
reliance
According
to
T
able
1,
these
studies
across
dif
ferent
cogniti
v
e
domains
indicate
that
understanding
cogniti
v
e
skills
in
children
with
ASD
in
v
olv
es
v
arious
le
v
els
and
types
of
ass
essment
depending
on
the
specic
features
of
each
cogniti
v
e
prole.
It
seems
that
each
child’
s
cogniti
v
e
and
de
v
elopmental
characteristics
are
Int
J
Artif
Intell,
V
ol.
15,
No.
4,
August
2026:
3164–3175
Evaluation Warning : The document was created with Spire.PDF for Python.
Int
J
Artif
Intell
ISSN:
2252-8938
❒
3167
reected
in
the
patterns
of
their
abilities,
and
naturally
,
the
most
prominent
analytical
methods
should
be
chosen
for
assessment
and
classication.
This
literature
analysis
sho
wed
that
understanding
and
measuring
the
cogniti
v
e
abilities
in
children
with
ASD
is
comple
x
and
not
straightforw
ard.
AI
algorithms
can
be
useful
in
identifying
correlations
between
cogniti
v
e
abilities
in
children
with
ASD
be
yond
traditional
statistical
methods.
In
addition,
adv
anced
methods,
such
as
clustering
and
f
actor
analysis,
can
be
ef
fecti
v
e
in
grouping
proles
within
the
ASD
data,
and
hence,
dene
more
accurate
correlations.
Deep
learning
can
also
be
useful
when
lar
ge
datasets
are
a
v
ailable
[31],
[32].
This
approach
can
help
adv
ance
our
understanding
of
ASD
abilities,
which
will
ha
v
e
impact
on
ho
w
to
handle
these
situations
at
earlier
stages.
3.
METHOD
This
study
aimed
to
e
xamine
the
relationships
between
cogniti
v
e
abilities
in
children
with
ASD.
A
structured
st
atistical
procedure
w
as
used
to
describe
the
data,
e
xamine
associations
between
v
ariables,
and
e
xplore
whether
the
cogniti
v
e
measures
formed
mea
n
i
ngful
subgroups
or
underlying
dimensions.
The
procedure
included:
i)
data
preparation
to
ensure
the
quality
and
consistenc
y
of
the
dataset,
ii)
descripti
v
e
statistical
analysis
to
characterise
the
main
features
of
the
v
ariables,
iii)
correlation
analysis
to
e
xamine
the
relationships
between
age
and
cogniti
v
e
functions,
i
v)
cluster
analysis
to
identify
possible
subgroups
within
the
sample,
and
v)
f
actor
analysis
to
in
v
estig
ate
latent
cogniti
v
e
structures
in
the
studied
population.
This
approach
w
as
used
to
pro
vide
an
account
of
cogniti
v
e
functioning
in
children
with
ASD,
while
allo
wing
for
both
common
patterns
across
the
sample
and
indi
vidual
dif
ferences
in
cogniti
v
e
proles.
3.1.
Data
pr
eparation
The
dataset
included
210
children
aged
6
to
12
years.
Six
v
ariables
were
included
in
the
analysi
s:
age,
VM,
FM,
SV
,
DM,
and
OR.
Children’
s
performance
w
as
assessed
through
v
e
cogniti
v
e
tasks
measuring
VM,
FM,
SV
,
DM,
and
OR.
Scores
for
each
cogniti
v
e
skill
ranged
from
1
to
20,
with
higher
v
alues
indicating
stronger
performance
on
the
corresponding
task.
As
sho
wn
in
T
able
2,
participant
1,
aged
8.2
years,
scored
6
in
VM,
7
in
FM,
11
in
SV
,
13
in
DM,
and
4
in
OR.
P
articipant
2,
aged
7.8
years,
scored
higher
in
VM
(8),
b
ut
lo
wer
in
FM
(4),
compared
with
participant
1.
P
articipant
3,
aged
7.0
years,
sho
wed
relati
v
ely
consistent
scores
across
VM,
FM,
and
SV
,
with
v
alues
of
5,
6,
and
6,
respecti
v
ely
,
while
sho
wing
higher
scores
in
DM
(14)
and
OR
(8).
These
e
xamples
illustrate
the
v
ariability
in
cogniti
v
e
performance
across
participants
and
across
cogniti
v
e
domains.
The
collected
data
were
therefore
suitable
for
e
xamining
patterns
of
cogniti
v
e
performance
across
age
and
skill
areas.
T
able
2
presents
a
sample
of
the
dataset.
T
able
2.
Cogniti
v
e
skill
measurements
for
participants
Seq.
Age
VM
FM
SV
DM
OR
1
8.2
6
7
11
13
4
2
7.8
8
4
6
13
7
3
7.0
5
6
6
14
8
4
7.6
8
11
11
12
8
5
8.5
6
9
12
6
8
6
12
6
5
5
5
9
3.2.
Descripti
v
e
statistical
analysis
Descripti
v
e
stat
istics
were
calculated
to
pro
vide
an
init
ial
characterization
of
the
sample
and
the
distrib
ution
of
the
cogniti
v
e
scores.
F
or
each
v
ariable,
the
mean,
standard
de
viation,
minimum,
maximum,
and
the
25th,
50th,
and
75th
percentiles
were
calculated.
These
statistics
were
use
d
to
describe
the
central
tendenc
y
and
v
ariability
of
the
data
before
conducting
further
analyses.
This
step
allo
wed
us
to
identify
whether
children
tended
to
score
higher
or
lo
wer
in
specic
cogniti
v
e
domains
and
whether
some
v
ariables
sho
wed
greater
dispersion
than
others.
F
or
e
xample,
VM
sho
wed
the
widest
range
of
scores,
while
FM
and
DM
sho
wed
moderate
v
ariability
.
This
pattern
suggested
that
the
sample
contained
considerable
heterogeneity
in
cogniti
v
e
abilities,
which
supported
the
need
for
additional
correlation,
clustering,
and
f
actor
analyses.
3.3.
Corr
elation
analysis
F
ollo
wing
the
descripti
v
e
analysis,
a
correlation
matrix
w
as
constructed
to
e
xamine
the
relationshi
ps
between
the
six
v
ariables.
Figure
1
sho
ws
the
tw
o
types
of
correlation
were
used:
Pearson
correlation
Exploring
co
gnitive
patterns
in
c
hildr
en
with
autism
spectrum
disor
der
using
corr
elation
and
...
(Hana
Bezzih)
Evaluation Warning : The document was created with Spire.PDF for Python.
3168
❒
ISSN:
2252-8938
for
linear
relationships
(Figure
1(a))
and
Spearman
correlation
for
rank-based
(monotonic)
relationships
(Figure
1(b)).
Pearson
correlation
measure
w
as
appl
ied
to
detect
the
strength
and
direction
of
linear
relationships
between
v
ariables.
The
correlation
coef
cient
ranges
from
-1
(perfect
ne
g
ati
v
e
correlation)
to
1
(perfect
positi
v
e
correlation),
with
v
alues
close
to
0
indicating
little
to
no
linear
relationship.
F
or
this
study
,
Pearson
correlations
were
used
to
assess
ho
w
well
one
cogniti
v
e
ability
can
af
fect
another
.
On
t
h
e
other
hand,
Spearman
correlation
is
useful
for
datasets
that
include
ordinal
or
rank-based
data.
Spearman
correlation
w
as
applied
as
a
non-parametric
alternati
v
e
to
Pearson.
Spearman
correlations
measure
ho
w
well
the
relationship
between
tw
o
v
ariables
can
be
described
using
a
monotonic
function,
without
assuming
a
linear
relationship.
The
correlation
matrices
pro
vided
insights
into
whether
certain
cogniti
v
e
abilities
were
related.
F
or
instance,
this
study
e
xplored
whether
children
who
scored
higher
in
VM
also
performed
well
in
FM
or
whether
age
inuenced
performance
in
specic
cogniti
v
e
domains.
(a)
(b)
Figure
1.
Correlation
matrices
for
all
v
ariables
of
(a)
Pearson
and
(b)
Spearman
3.4.
Cluster
analysis
Ne
xt,
cluster
analysi
s
w
as
performed
to
gr
o
up
the
childre
n
int
o
dist
inct
clusters
based
on
their
cogniti
v
e
abilities.
PCA
is
used
to
process
v
ariables
and
K-means
algorithm
then
partitions
the
dataset
into
k
clusters,
where
k
is
a
predened
number
of
groups,
and
each
child
is
assigned
to
the
cluster
with
the
nearest
mean.
K-means
algorithm
is
used
to
create
clusters.
This
algorithm
seeks
to
minimize
the
within-cluster
v
ariance
by
adjusting
cluster
centroids
iterati
v
ely
as
illustrated
in
Figure
2.
–
Cluster
selection:
a
k
e
y
part
of
the
K-means
process
is
determining
the
appropriate
number
of
clusters.
This
study
initially
set
k
=
3
,
suggesting
that
children
with
ASD
might
be
grouped
into
clusters
of
high,
moderate,
and
lo
w
performance
across
cogniti
v
e
d
om
ains.
Cluster
v
alidity
w
as
then
e
xamined
using
the
silhouette
score,
which
e
v
aluates
the
quality
of
clustering
by
meas
uring
ho
w
similar
each
data
point
is
to
its
o
wn
cluster
compared
to
other
clusters.
Scores
close
to
1
indicate
well-dened
clusters,
wherea
s
scores
near
0
indicate
o
v
erlap
between
clusters.
–
Cluster
characterization:
after
clustering,
the
mean
v
alues
of
the
six
v
ariables
within
each
cluster
were
e
xamined
to
characterize
the
cogniti
v
e
proles
of
children
in
each
group.
This
can
sho
w
if
children
in
certain
clusters
tended
to
ha
v
e
high
performance
in
specic
cogniti
v
e
areas,
or
displayed
more
generalized
tendenc
y
across
dif
ferent
domains,
for
instance
stronger
or
weak
er
correlation.
3.5.
F
actor
analysis
F
actor
analysis
is
a
v
ariables
reduction
method
that
combines
correlated
v
ariables
together
into
f
actors,
in
order
to
pro
vide
more
focus
on
ho
w
these
v
ariables
together
inuence
the
outcomes.
F
or
f
actor
e
xtraction,
PCA
w
as
applied
for
identi
fying
potential
f
actors
where
v
ariables
are
e
xpected
to
be
correlated.
Then,
after
the
f
actors
were
e
xtracted,
rotation
method
w
as
applied
to
clarify
which
v
ariables
contrib
ute
the
most
to
each
Int
J
Artif
Intell,
V
ol.
15,
No.
4,
August
2026:
3164–3175
Evaluation Warning : The document was created with Spire.PDF for Python.
Int
J
Artif
Intell
ISSN:
2252-8938
❒
3169
f
actor
.
Finally
,
f
actor
loa
d
i
ngs,
which
represent
the
correlations
between
the
original
v
ariables
and
the
e
xtracted
f
actors,
were
used
to
dene
the
cogniti
v
e
structures.
F
or
instance,
a
high
loading
of
VM
and
FM
on
the
same
f
actor
indicates
that
these
tw
o
abilities
tend
to
correlate.
The
methodology
in
v
olv
ed
a
multi-step
process
designed
to
capture
the
relationships
between
cogniti
v
e
v
ariables.
The
results
of
the
correlation,
clustering,
and
f
actor
analysis
were
used
to
pro
vide
the
de
gree
of
linear
and
rank-based
relationships
between
the
cogniti
v
e
abilities.
While
cluster
analysis
can
t
cogniti
v
e
proles
within
the
dataset
into
groups,
f
actor
analysis
can
unco
v
er
hidden
cogniti
v
e
dimensions
that
could
e
xplain
ho
w
these
abilities
were
related
or
distinct.
Figure
2.
Clusters
of
v
ariables
using
PCA
and
K-means
4.
RESUL
TS
AND
AN
AL
YSIS
The
general
statistics
pro
vided
information
about
the
cogniti
v
e
abilities
of
the
children.
The
a
v
erage
age
w
as
approximately
9.14
years,
with
a
standard
de
viation
of
1.32.
The
cogniti
v
e
measures
for
all
v
e
abilities
under
test
sho
wed
mean
scores
generally
in
the
“at
risk”
range
of
scores
between
8
and
12.
VM
had
the
widest
range
of
scores,
from
2
to
19,
while
other
v
ariables
sho
wed
moderate
v
ar
iability
.
In
the
rst
step,
a
correlation
matrix
analysis
w
as
conducted,
which
re
v
ealed
weak
relationships
between
most
v
ariables,
as
sho
wn
in
Figure
1.
Age
did
not
signi
cantly
correlate
with
an
y
cogniti
v
e
measures,
suggesting
that
within
this
age
group,
age
w
as
not
a
strong
predictor
of
performance
on
the
assessed
tasks.
A
moderate
positi
v
e
correlation
between
VM
and
FM
(0.34)
indicated
that
c
h
i
ldren
who
performed
better
in
visual
motor
tasks
tended
to
score
higher
in
FM.
A
weak
ne
g
ati
v
e
correlation
emer
ged
between
DM
and
OR
(
−
0
.
19
),
pointing
to
a
slight
in
v
erse
relationship
between
these
abilities.
The
most
notable
relationship
observ
ed
w
as
a
moderate
positi
v
e
correlation
(0.34)
between
VM
and
FM,
suggesting
that
these
skills
may
share
some
underlying
cogniti
v
e
mechanisms.
Age
sho
wed
ne
gligible
correlations
with
all
cogniti
v
e
measures,
indicating
that
it
did
not
signicantly
inuence
performance
on
these
tasks
in
this
dataset.
SV
demonstrated
a
moderate
positi
v
e
correlation
(0.183)
with
FM,
while
other
correlations
between
v
ariables
were
generally
weak
or
ne
gligible.
DM
appeared
lar
gely
independent
of
other
cogniti
v
e
abilities,
with
v
ery
weak
correlations
across
the
board.
T
ak
en
together
,
the
analysis
suggests
that
while
some
cogniti
v
e
tasks
such
as
VM
and
FM
may
be
some
what
related,
most
of
the
measured
abilities
operate
with
relati
v
e
independence
from
each
other
.
A
cluster
analysis
w
as
conducted,
which
grouped
the
dat
a
into
three
clusters,
with
the
lar
gest
clust
er
ha
ving
78
data
points
and
the
smallest
ha
ving
61.
Ho
we
v
er
,
the
silhouette
score
w
as
lo
w
(0.1368),
indicating
poor
s
eparation
between
clusters.
This
suggests
that
the
v
ariables
used,
namely
age,
VM,
FM,
SV
,
DM,
and
OR,
may
not
clearly
distinguish
the
groups.
The
weak
correlations
within
clusters,
coupled
with
the
lo
w
silhouette
score,
imply
potential
o
v
erlap
in
cogniti
v
e
proles
among
children
across
dif
ferent
clusters,
mainly
in
terms
of
linear
correlation.
Figure
3
sho
ws
correlation
between
all
v
ariables
for
cluster
0.
There
are
some
small
positi
v
e
correlations
between
v
ariables,
such
as
between
age
and
VM
(0.12)
and
between
FM
and
DM
(0.20).
Ho
we
v
er
,
these
correlations
are
generally
weak.
SV
and
OR
sho
w
notable
ne
g
ati
v
e
correlations
with
other
v
ariables,
such
as
-0.17
with
age
and
-0.24
with
SV
.
This
indicates
that
as
these
v
ariables
increase,
others
tend
to
decrease.
The
heatmap
indicates
that
there
are
no
v
ery
strong
correlations
am
on
g
v
ariables
within
Exploring
co
gnitive
patterns
in
c
hildr
en
with
autism
spectrum
disor
der
using
corr
elation
and
...
(Hana
Bezzih)
Evaluation Warning : The document was created with Spire.PDF for Python.
3170
❒
ISSN:
2252-8938
cluster
0.
Most
correl
ations
are
weak,
with
a
fe
w
moderate
ones.
This
suggests
that
the
v
ariables
do
not
ha
v
e
strong
linear
relationships
within
this
cluster
.
Figure
4
also
sho
ws
correlation
between
all
v
ariables
for
cluster
1.
The
most
notable
relationship
is
the
positi
v
e
correlation
between
age
and
DM
(0.440),
suggesti
ng
that
age
has
a
more
substantial
ef
fect
on
DM
compared
to
other
v
ariables.
SV
tends
to
sho
w
more
signicant
ne
g
ati
v
e
correlations
with
VM,
indicating
an
in
v
erse
relationship
in
this
cluster
.
Other
correlations
are
generally
weak,
with
man
y
v
alues
close
to
zero,
indicating
minim
al
relationships
between
those
v
ariables.
Figure
5
sho
ws
correlation
between
all
v
ariables
for
cluster
2.
The
most
notable
relationship
in
this
cluster
is
the
positi
v
e
correlation
between
age
and
DM
(0.304),
suggesting
that
age
has
a
more
signicant
ef
fect
on
DM
in
this
cluster
.
VM
and
SV
sho
w
a
moderate
ne
g
ati
v
e
correlation,
indicating
an
in
v
erse
relationship.
Most
other
correlations
are
weak,
with
se
v
eral
v
ariables
ha
ving
minimal
or
no
signicant
relationships
with
each
other
.
Figure
3.
Correlation
matrix
between
all
v
ariables
for
clusters
0
Figure
4.
Correlation
matrix
between
all
v
ariables
for
clusters
1
Figure
6
sho
ws
heatmap
for
relationships
between
v
ariables
and
f
actors.
VM
(0.603)
and
FM
(0.642)
ha
v
e
strong
positi
v
e
loadings
on
f
actor
1,
suggesting
that
these
v
ariables
are
closely
related
and
lik
ely
contrib
ute
Int
J
Artif
Intell,
V
ol.
15,
No.
4,
August
2026:
3164–3175
Evaluation Warning : The document was created with Spire.PDF for Python.
Int
J
Artif
Intell
ISSN:
2252-8938
❒
3171
to
this
f
actor
.
SV
(0.329)
also
has
a
moderate
positi
v
e
loading,
indicating
it
shares
some
relationship
with
this
f
actor
b
ut
is
less
strongly
associated.
Age
(-0.088)
and
DM
(0.040)
ha
v
e
v
ery
lo
w
loadings,
suggesting
the
y
do
not
contrib
ute
signicantly
to
f
actor
1.
OR
(0.235)
demonstrated
a
weak
positi
v
e
loading,
indicating
a
limited
association
with
f
actor
1.
F
actor
1
w
as
strongly
associated
with
VM
and
FM,
suggesting
that
these
tw
o
v
ariables
represent
a
shared
cogniti
v
e
dimension.
The
gure
also
illustrates
that
DM
(-0.404)
demonstrated
a
st
rong
ne
g
ati
v
e
loading
on
f
actor
2,
indicating
an
in
v
erse
relationship
with
this
f
actor
.
OR
(0.372)
sho
wed
a
moderate
positi
v
e
loading,
reecting
a
moderate
association
with
f
actor
2.
VM
(0.018),
FM
(-0.097),
and
age
(0.010)
demonstrated
v
ery
lo
w
loadings,
suggesting
that
these
v
ariables
do
not
signicantly
contrib
ute
to
this
f
actor
.
SV
(-0.003)
e
xhibited
a
ne
gligible
loading,
indicating
the
absence
of
a
signicant
relationship
with
f
actor
2.
Therefore,
f
actor
2
demonstrated
a
strong
ne
g
ati
v
e
association
with
DM
and
a
moderate
positi
v
e
association
with
OR,
suggesting
that
these
abilities
are
connected
to
a
dif
ferent
cogniti
v
e
dimension.
Figure
5.
Correlation
matrix
between
all
v
ariables
for
cluster
2
Figure
6.
Relationships
between
v
ariables
and
f
actors
Exploring
co
gnitive
patterns
in
c
hildr
en
with
autism
spectrum
disor
der
using
corr
elation
and
...
(Hana
Bezzih)
Evaluation Warning : The document was created with Spire.PDF for Python.
3172
❒
ISSN:
2252-8938
Finally
,
SV
(0.357)
demonst
rated
a
moderate
positi
v
e
loading
on
f
actor
3,
indicating
a
notable
relationship
with
this
f
actor
.
VM
(-0.142)
and
DM
(-0.167)
e
xhibited
weak
ne
g
ati
v
e
loadings,
suggesting
limited
relationships
with
this
f
actor
,
while
FM
(0.025)
and
OR
(-0.153)
ha
v
e
v
ery
weak
loadings,
indicating
minimal
contrib
ution
to
f
actor
3.
Age
(-0.127)
also
has
a
v
ery
weak
loading.
Thus,
f
actor
3
w
as
mainly
dri
v
en
by
SV
,
with
minimal
contrib
utions
from
other
v
ariables.
These
results
deserv
e
attention
because
the
y
suggest
that
the
underlying
cogniti
v
e
structure
is
characterized
by
distinct
dimensions,
with
some
v
ariables
clustering
together
while
others
sho
w
contrasting
relationships.
It
seems
that
each
f
actor’
s
cogniti
v
e
composition
is
reected
in
the
de
v
elopment
of
the
child’
s
processing
frame
w
ork,
and
naturally
,
the
most
prominent
dimension
captures
the
abilities
most
closely
associated
with
shared
cogniti
v
e
mechanisms.
Figure
7
pro
vides
a
comprehensi
v
e
vie
w
of
the
data
f
actor
scores
and
loadings.
Figure
7.
Comprehensi
v
e
vie
w
of
the
data
f
actor
scores
and
loadings
5.
DISCUSSION
AND
RECOMMEND
A
TIONS
In
the
present
analysis,
weak
correlations
were
found
among
most
of
the
assessed
abilities,
suggesting
that
these
cogniti
v
e
domains
operate
with
relati
v
e
independence.
This
nding
challenges
the
assumption
that
cogniti
v
e
skills
in
children
with
ASD
are
closely
connected
or
de
v
elop
in
a
uniform
pattern.
Instead,
it
supports
the
vie
w
that
cogniti
v
e
de
v
elopment
in
ASD
is
comple
x,
une
v
en,
and
highly
indi
vidual
during
this
important
age
range.
The
analysis
of
cogniti
v
e
skills
in
children
with
ASD
of
fers
se
v
eral
practical
benets.
By
identifying
specic
cogniti
v
e
proles,
interv
entions
can
be
personalized
and
made
more
responsi
v
e
to
the
child’
s
strengths
and
weaknesses.
Distinct
patterns
of
cogniti
v
e
functioning
may
also
support
earlier
ASD
identication
and
more
timely
interv
ention.
These
ndings
may
also
inform
teaching
strate
gies
that
are
better
aligned
with
the
cogniti
v
e
processing
styles
of
children
with
ASD
[33].
The
inte
gration
of
PCA-dri
v
en
clustering
with
f
actor
analysis
in
this
study
pro
vi
d
e
s
a
useful
frame
w
ork
for
identifying
latent
cogniti
v
e
structures
in
ASD
populations.
This
multi-method
approach
allo
wed
us
to
e
xamine
both
possible
subgroups
and
underlying
dimensions
of
cogniti
v
e
v
ariability
.
The
ndings
suggest
potential
applications
in
adapti
v
e
educational
tools
and
personalised
therap
y
plans.
This
may
be
useful
because
the
moderate
positi
v
e
correlation
between
VM
and
FM
suggests
some
shared
cogniti
v
e
mechanisms
between
these
abilities.
Future
e
xtensions
may
include
AI-enhanced
clustering
methods
that
use
deep
learning
algorithms
for
more
sophisticated
pattern
recognition,
longitudinal
tracking
systems
that
monitor
cogniti
v
e
de
v
elopment
o
v
er
e
xtended
periods,
and
multimodal
inte
gration
of
beha
vioral,
cogniti
v
e,
and
neurobiological
data.
ML
prediction
models
may
also
be
de
v
eloped
to
forecast
de
v
elopmental
trajectories
from
early
cogniti
v
e
proles,
ena
b
l
ing
ea
rlier
and
m
ore
ta
r
get
ed
i
nterv
ention.
In
addition,
adapti
v
e
assessment
tools
that
adjust
in
real
time
according
to
a
child’
s
cogniti
v
e
prole
may
pro
vide
more
accurate
and
indi
vidualised
e
v
aluation.
Int
J
Artif
Intell,
V
ol.
15,
No.
4,
August
2026:
3164–3175
Evaluation Warning : The document was created with Spire.PDF for Python.
Int
J
Artif
Intell
ISSN:
2252-8938
❒
3173
Finally
,
this
study
intends
to
consider
other
aspects
in
this
group
of
children,
including
attention
decit
h
yperacti
vity
disorder
[34].
6.
CONCLUSION
This
study
in
v
estig
ated
cogniti
v
e
patterns
in
children
with
ASD
aged
6
to
12
years
using
correl
ation
analysis,
cluster
analysis,
and
f
actor
analysis.
The
results
sho
wed
weak
correlations
between
most
cogniti
v
e
v
ariables,
and
age
w
as
not
signicantly
associated
with
cogniti
v
e
performance.
The
clearest
association
w
as
a
moderate
positi
v
e
correlation
(0.34)
between
VM
and
FM,
suggesting
possible
shared
underlying
mechanisms.
Cluster
analysis
identied
three
groups,
b
ut
the
lo
w
silhouette
score
(0.13
68)
indicated
weak
separation
and
substantial
o
v
erlap
between
cogniti
v
e
proles.
F
act
or
analysis
identied
three
dimensions:
f
actor
1
w
as
associated
with
VM
and
FM,
f
actor
2
w
as
associated
with
DM
and
OR,
and
f
actor
3
w
as
primarily
associated
with
SV
.
These
ndings
demonstrate
the
heterogeneous
nature
of
cogniti
v
e
abilities
in
children
with
ASD
and
support
the
need
for
indi
vidualized
assessment
and
i
nterv
ention.
The
study
contrib
utes
a
frame
w
ork
that
combines
PCA-dri
v
en
clustering
with
f
actor
analysis
for
cogniti
v
e
proling
in
ASD.
Future
research
should
e
xamine
longitudinal
de
v
elopmental
trajectories,
AI-enhanced
prediction
models,
broader
cogniti
v
e
domains,
and
the
inte
gration
of
neurobiological
and
en
vironmental
f
actors.
Exploring
non-linear
relationships
and
using
more
sophisticated
clustering
techniques
may
re
v
eal
patterns
that
were
not
captured
in
the
present
analysis.
Combining
cogniti
v
e
assessment
with
neuro-imaging
may
also
help
link
cogniti
v
e
proles
to
underlying
brain
structure
and
function.
FUNDING
INFORMA
TION
Authors
state
no
funding
in
v
olv
ed.
A
UTHOR
CONTRIB
UTIONS
ST
A
TEMENT
This
journal
uses
the
Contrib
utor
Roles
T
axonomy
(CRediT)
to
recognize
indi
vidual
author
contrib
utions,
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
Hana
Bezzih
✓
✓
✓
✓
✓
✓
Muna
Darweesh
✓
✓
✓
✓
✓
✓
✓
Amjad
Ga
w
anmeh
✓
✓
✓
✓
✓
✓
✓
✓
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
Authors
state
no
conict
of
interest.
D
A
T
A
A
V
AILABILITY
The
data
that
support
the
ndings
of
this
study
are
a
v
ailable
on
request
from
the
corresponding
author
,
[A
G].
The
data,
which
contain
information
that
could
compromise
the
pri
v
ac
y
of
research
participants,
are
not
publicly
a
v
ailable
due
to
certain
restrictions.
REFERENCES
[1]
M.
Stanti
´
c,
E.
Ichijo,
C.
Catmur
,
and
G.
Bird,
“F
ace
memory
and
f
ace
perception
in
autism,
”
A
utism
,
v
ol.
26,
no.
1,
pp.
276–280,
Jan.
2022,
doi:
10.1177/13623613211027685.
[2]
J.
A.
Little,
“V
isi
on
in
children
with
autism
spectrum
disorder:
a
critical
re
vie
w
,
”
Clinical
and
e
xperimental
opt
ometry
,
v
ol.
101,
no.
4,
pp.
504–513,
Jul.
2018,
doi:
10.1111/cxo.12651.
Exploring
co
gnitive
patterns
in
c
hildr
en
with
autism
spectrum
disor
der
using
corr
elation
and
...
(Hana
Bezzih)
Evaluation Warning : The document was created with Spire.PDF for Python.