I
AE
S
I
n
t
e
r
n
at
ion
al
Jou
r
n
al
of
Ar
t
if
icial
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
AI
)
Vol.
14
,
No.
4
,
Augus
t
2025
,
pp.
3033
~
3046
I
S
S
N:
2252
-
8938
,
DO
I
:
10
.
11591/i
jai
.
v
14
.i
4
.
pp
30
33
-
3046
3033
Jou
r
n
al
h
omepage
:
ht
tp:
//
ij
ai
.
iaes
c
or
e
.
c
om
M
ac
h
i
n
e
l
e
ar
n
in
g f
o
r
gl
ob
al
t
r
ad
e
a
n
al
ysi
s:
a h
y
b
r
id
c
lu
st
e
r
in
g
ap
p
r
oac
h
u
si
n
g D
B
S
C
A
N
,
e
lb
ow
, an
d
S
OM
M
u
s
d
ali
f
a
T
h
am
r
in
1
,
I
d
a
M
u
lyad
i
2
,
I
De
wa
M
a
d
e
Wid
ia
3
,
M
u
h
am
m
ad
F
ais
al
2
,
S
u
ar
d
i
Hi
B
ah
a
r
u
d
d
in
1
,
M
e
d
y
Wis
n
u
P
r
ih
a
t
m
on
o
4
,
Nurdi
an
s
yah
5
,
Nas
ir
Us
m
an
4
1
D
e
pa
r
tm
e
nt
of
I
nf
or
ma
ti
c
s
M
a
na
ge
me
nt
, S
T
M
I
K
P
r
of
e
s
io
na
l
M
a
ka
s
s
a
r
, M
a
k
a
s
s
a
r
, I
ndone
s
ia
2
D
e
pa
r
tm
e
nt
of
I
nf
or
ma
ti
c
s
, U
ni
ve
r
s
it
a
s
M
uha
mm
a
di
ya
h M
a
k
a
s
s
a
r
, M
a
k
a
s
s
a
r
, I
ndone
s
ia
3
F
a
c
ul
ty
of
V
oc
a
ti
ona
l
S
tu
di
e
s
,
U
ni
ve
r
s
it
a
s
B
r
a
w
ij
a
ya
,
M
a
la
ng
,
I
ndone
s
ia
4
D
e
pa
r
tm
e
nt
of
I
nf
or
ma
ti
on S
ys
te
m, S
T
M
I
K
P
r
of
e
s
io
na
l
M
a
ka
s
s
a
r
, M
a
k
a
s
s
a
r
, I
ndone
s
ia
5
D
e
pa
r
tm
e
nt
of
D
ig
it
a
l
B
us
in
e
s
s
, U
ni
ve
r
s
it
a
s
D
ip
a
M
a
ka
s
s
a
r
,
M
a
ka
s
s
a
r
, I
ndone
s
ia
Ar
t
icle
I
n
f
o
AB
S
T
RA
CT
A
r
ti
c
le
h
is
tor
y
:
R
e
c
e
ived
De
c
4,
2024
R
e
vis
e
d
J
un
14,
2025
Ac
c
e
pted
J
ul
10,
2025
G
l
o
b
a
l
t
rad
e
co
n
s
t
i
t
u
t
e
s
a
h
i
g
h
l
y
co
m
p
l
e
x
an
d
i
n
t
erd
ep
e
n
d
e
n
t
s
y
s
t
em
i
n
f
l
u
e
n
ced
b
y
d
i
v
er
s
e
eco
n
o
mi
c,
g
e
o
g
ra
p
h
i
c,
an
d
p
o
l
i
t
i
c
al
fact
o
r
s
.
T
h
i
s
s
t
u
d
y
p
ro
p
o
s
es
a
h
y
b
ri
d
c
l
u
s
t
er
i
n
g
framew
o
r
k
t
h
at
i
n
t
eg
ra
t
es
d
en
s
i
t
y
-
b
as
e
d
s
p
at
i
al
cl
u
s
t
er
i
n
g
o
f
ap
p
l
i
ca
t
i
o
n
s
w
i
t
h
n
o
i
s
e
(D
BSC
A
N
),
el
b
o
w
,
an
d
s
el
f
-
o
r
g
an
i
zi
n
g
map
s
(SO
M)
me
t
h
o
d
s
t
o
u
n
co
v
er
l
a
t
en
t
s
t
ru
c
t
u
re
s
i
n
i
n
t
ern
a
t
i
o
n
a
l
t
ra
d
e
p
at
t
ern
s
.
U
t
i
l
i
zi
n
g
av
era
g
ed
t
rad
e
d
at
a
fr
o
m
2
5
c
o
u
n
t
r
i
es
s
p
a
n
n
i
n
g
t
h
e
p
er
i
o
d
fro
m
2
0
1
3
t
o
2
0
2
3
,
t
h
e
framew
o
rk
i
d
en
t
i
f
i
es
d
i
s
t
i
n
c
t
cl
u
s
t
ers
b
as
e
d
o
n
ex
p
o
rt
-
i
mp
o
rt
c
h
aract
er
i
s
t
i
c
s
.
T
h
e
D
BSCA
N
i
s
emp
l
o
y
ed
t
o
d
et
ec
t
d
en
s
e
t
rad
e
h
u
b
s
an
d
o
u
t
l
i
er
b
eh
a
v
i
o
rs
,
t
h
e
el
b
o
w
met
h
o
d
d
e
t
ermi
n
es
t
h
e
o
p
t
i
mal
n
u
mb
er
o
f
cl
u
s
t
er
s
,
an
d
SO
M
faci
l
i
t
a
t
es
t
h
e
v
i
s
u
a
l
i
za
t
i
o
n
o
f
n
o
n
-
l
i
n
ear,
h
i
g
h
-
d
i
me
n
s
i
o
n
al
t
ra
d
e
rel
at
i
o
n
s
h
i
p
s
.
T
h
e
an
al
y
s
i
s
rev
ea
l
s
t
h
ree
p
ro
m
i
n
e
n
t
t
ra
d
e
cl
u
s
t
er
s
:
G
l
o
b
al
T
rad
e
L
ead
ers
,
E
merg
i
n
g
T
ra
d
e
Po
w
er
s
,
an
d
N
i
c
h
e
E
x
p
o
r
t
ers
,
each
refl
ect
i
n
g
v
ary
i
n
g
d
e
g
rees
o
f
t
rad
e
d
i
v
ers
i
fi
ca
t
i
o
n
an
d
d
ep
e
n
d
e
n
cy
.
T
h
e
s
e
emp
i
ri
cal
fi
n
d
i
n
g
s
al
i
g
n
w
i
t
h
e
s
t
a
b
l
i
s
h
ed
eco
n
o
mi
c
t
h
e
o
ri
e
s
,
i
n
c
l
u
d
i
n
g
t
h
e
H
eck
s
ch
er
O
h
l
i
n
mo
d
el
a
n
d
d
ep
en
d
e
n
cy
t
h
e
o
ry
,
an
d
p
ro
v
i
d
e
act
i
o
n
ab
l
e
i
n
s
i
g
h
t
s
f
o
r
p
o
l
i
cy
ma
k
ers
s
eek
i
n
g
t
o
en
h
an
ce
t
ra
d
e
co
mp
e
t
i
t
i
v
en
e
s
s
an
d
reg
i
o
n
al
i
n
t
e
g
rat
i
o
n
s
t
rat
e
g
i
e
s
.
K
e
y
w
o
r
d
s
:
C
lus
ter
ing
DB
S
C
AN
E
lbow
S
e
lf
-
or
ga
nizing
maps
T
r
a
de
pa
tt
e
r
ns
Th
i
s
i
s
a
n
o
p
en
a
c
ces
s
a
r
t
i
c
l
e
u
n
d
e
r
t
h
e
CC
B
Y
-
SA
l
i
ce
n
s
e.
C
or
r
e
s
pon
din
g
A
u
th
or
:
M
us
da
li
f
a
T
ha
mr
in
De
pa
r
tm
e
nt
of
I
nf
or
mat
ics
M
a
na
ge
ment,
S
T
M
I
K
P
r
of
e
s
ional
M
a
ka
s
s
a
r
J
L
.
AP
P
e
t
t
a
r
a
ni
S
tr
e
e
t,
No.
27
,
P
a
na
kuka
ng,
M
a
ka
s
s
a
r
,
I
ndone
s
ia
E
mail:
nonongtham
r
in@gm
a
il
.
c
om
1.
I
NT
RODU
C
T
I
ON
Globa
l
tr
a
de
ha
s
incr
e
a
s
in
gly
be
e
n
r
e
c
ogniz
e
d
a
s
a
c
ompl
e
x,
mul
ti
f
a
c
e
ted
phe
nomenon
with
pr
of
ound
i
mpl
ica
ti
ons
f
o
r
the
e
c
onomi
c
lands
c
a
pe
s
of
na
ti
ons
[
1]
.
C
onve
nti
ona
l
methods
of
tr
a
de
a
na
lys
is
,
f
r
e
que
ntl
y
f
oc
us
e
d
on
f
inanc
ial
indi
c
a
tor
s
a
nd
bil
a
ter
a
l
tr
a
de
models
,
may
f
a
il
to
c
a
ptur
e
the
int
r
i
c
a
te
a
nd
dyna
mi
c
na
tur
e
of
int
e
r
na
ti
ona
l
t
r
a
de
f
lows
.
R
e
c
e
nt
a
dva
nc
e
ment
s
in
da
ta
a
va
il
a
bil
it
y
a
nd
a
na
lyt
ica
l
tec
hniques
,
pa
r
ti
c
ular
ly
thr
ough
c
lus
ter
ing
meth
ods
,
a
nd
pr
e
s
e
nt
ne
w
a
ve
nue
s
f
or
c
ompr
e
he
nding
thes
e
dyna
mi
c
s
by
e
na
bli
ng
the
gr
oup
ing
of
c
ountr
ies
ba
s
e
d
on
s
im
il
a
r
it
ies
in
thei
r
e
xpor
t
a
nd
im
po
r
t
p
r
of
il
e
s
[
2]
.
P
r
ior
a
na
lys
e
s
ha
ve
unde
r
s
c
or
e
d
the
va
lue
of
s
ophis
ti
c
a
ted
methodologi
e
s
in
e
nha
nc
ing
the
pr
e
dictive
powe
r
a
nd
in
ter
pr
e
ti
ve
c
a
pa
bil
it
ies
o
f
tr
a
de
s
tudi
e
s
.
R
obus
t
r
e
g
r
e
s
s
ion
a
ppr
oa
c
he
s
by
Nif
ti
ye
v
[
3]
a
nd
ti
me
s
e
r
ies
a
na
lys
e
s
by
Alz
a
hr
a
ni
a
nd
S
a
lah
[
4]
il
lus
tr
a
te
the
e
f
f
ica
c
y
o
f
e
mpl
oying
a
dva
nc
e
d
a
na
lyt
ica
l
tec
hniques
in
int
e
r
na
ti
ona
l
tr
a
de
c
ontexts
.
T
he
s
e
s
tudi
e
s
unde
r
s
c
or
e
the
ne
e
d
to
move
be
yond
tr
a
dit
ional
e
c
onometr
ic
models
to
ga
in
a
de
e
pe
r
unde
r
s
tanding
of
tr
a
de
dyna
m
ics
.
Additi
ona
ll
y
,
qu
a
li
tative
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
20
25
:
303
3
-
3046
3034
methodologi
e
s
a
nd
text
-
ba
s
e
d
a
na
lys
e
s
,
a
s
de
mons
tr
a
ted
in
[
5]
,
[
6
]
,
pr
ovide
va
luable
c
ompl
e
menta
r
y
ins
ight
s
,
pa
r
ti
c
ular
ly
in
e
va
luating
the
im
pa
c
ts
of
ge
opoli
ti
c
a
l
e
ve
nts
s
uc
h
a
s
the
2022
e
s
c
a
lation
of
the
R
us
s
ia
-
Ukr
a
ine
wa
r
.
I
n
tegr
a
ti
ng
thes
e
a
l
ter
na
ti
ve
a
pp
r
oa
c
he
s
c
a
n
s
igni
f
ica
ntl
y
e
nr
ich
e
c
onometr
ic
a
nd
s
tatis
ti
c
a
l
e
va
lu
a
ti
ons
.
C
lus
ter
ing
tec
hniques
,
a
s
a
s
ubs
e
t
of
uns
upe
r
vis
e
d
mac
hine
lea
r
ning,
f
a
c
il
it
a
te
the
identif
ica
ti
on
of
late
nt
gr
oupings
withi
n
tr
a
de
da
ta,
potentially
unve
il
ing
hidden
s
t
r
uc
tur
e
s
a
nd
r
e
lations
hips
that
e
lude
tr
a
dit
ional
models
.
F
o
r
ins
tanc
e
,
r
e
s
e
a
r
c
h
ha
s
de
m
ons
tr
a
ted
that
tr
a
de
ne
two
r
ks
c
a
n
e
xhibi
t
no
table
s
tr
uc
tur
a
l
c
ha
r
a
c
ter
is
ti
c
s
with
the
c
a
pa
c
it
y
to
inf
luenc
e
t
r
a
de
f
lows
s
igni
f
ica
ntl
y.
T
he
vi
r
tual
wa
ter
tr
a
de
n
e
twor
ks
a
na
lyze
d
by
Xing
a
nd
C
he
n
[
7
]
p
r
ovide
a
r
e
leva
nt
e
xa
mpl
e
,
il
lus
tr
a
ti
ng
how
tr
a
de
pa
tt
e
r
ns
c
a
n
s
ubs
t
a
nti
a
ll
y
a
f
f
e
c
t
the
dis
tr
ibut
ion
of
wa
ter
r
e
s
our
c
e
s
a
c
r
os
s
na
ti
ons
.
He
r
z
be
r
ge
r
e
t
al
.
[
8]
s
im
il
a
r
ly
unde
r
s
c
or
e
the
im
por
tanc
e
of
t
r
a
de
r
e
lations
hips
,
e
mphas
izing
t
ha
t
int
e
r
a
c
ti
ons
a
c
r
os
s
va
r
ious
tr
a
de
s
ys
tems
may
e
it
he
r
e
nha
nc
e
or
c
ounter
ba
lanc
e
tr
a
de
f
lows
,
ther
e
by
c
ompl
ica
ti
ng
e
f
f
o
r
t
s
to
unde
r
s
tand
the
mul
ti
f
a
c
e
ted
na
tur
e
of
global
tr
a
de
dyna
mi
c
s
.
M
or
e
ove
r
,
a
pplyi
ng
mac
hine
lea
r
ning
methodolog
ies
in
tr
a
de
f
or
e
c
a
s
ti
ng
ha
s
pr
ove
n
a
dva
ntage
ous
in
r
e
f
ini
ng
pr
e
dictive
a
c
c
ur
a
c
y.
J
oš
ić
a
nd
Ž
muk
[
9]
a
r
gue
that
mac
hine
lea
r
ning
a
lgor
it
hms
of
f
e
r
c
r
uc
ial
ins
ight
s
f
or
poli
c
ymake
r
s
a
nd
r
e
s
e
a
r
c
he
r
s
by
e
n
ha
nc
ing
the
pr
e
c
is
ion
of
bil
a
ter
a
l
tr
a
de
f
low
pr
e
dictions
.
Gopina
th
e
t
al
.
[
10
]
c
or
r
obo
r
a
te
th
e
s
e
f
indi
ngs
,
de
mons
tr
a
ti
ng
that
mac
hine
lea
r
ning
tec
hniqu
e
s
hold
pa
r
ti
c
ular
e
f
f
ica
c
y
in
f
or
e
c
a
s
ti
ng
a
gr
icultu
r
a
l
tr
a
de
,
yielding
s
upe
r
ior
long
-
ter
m
f
it
s
whe
n
c
om
pa
r
e
d
to
tr
a
dit
ional
e
c
onometr
ic
models
.
T
hus
,
int
e
gr
a
ti
ng
a
dva
nc
e
d
a
na
lyt
ica
l
f
r
a
mew
or
ks
int
o
tr
a
de
a
na
lys
is
c
ontr
ibut
e
s
to
a
mo
r
e
nua
nc
e
d
unde
r
s
tanding
o
f
gl
oba
l
tr
a
de
pa
tt
e
r
ns
.
T
he
c
lus
ter
ing
of
c
ountr
ies
ba
s
e
d
on
tr
a
de
p
r
of
il
e
s
a
ddit
ionally
p
r
ovides
e
s
s
e
nti
a
l
ins
ight
s
int
o
the
e
c
onomi
c
int
e
r
de
pe
nde
nc
ies
that
c
ha
r
a
c
ter
i
z
e
c
on
tempor
a
r
y
tr
a
de
r
e
la
ti
ons
hips
.
W
hil
e
the
gr
a
vit
y
model
of
tr
a
de
,
a
n
e
xtens
ively
e
mpl
oye
d
tool
in
tr
a
de
f
lo
w
a
na
lys
is
,
s
ugge
s
t
s
that
e
c
onomi
c
s
ize
a
nd
g
e
ogr
a
phic
dis
tanc
e
a
r
e
c
r
it
ica
l
de
ter
mi
na
nts
o
f
bi
late
r
a
l
t
r
a
de
[
11]
,
R
a
s
ouli
ne
z
ha
d
a
nd
J
a
ba
lame
li
[
11]
c
ontend
that
the
c
ompl
e
x
int
e
gr
a
ti
on
pa
tt
e
r
ns
a
mong
B
r
a
z
il
,
R
us
s
ia,
I
ndia,
a
nd
C
hina
(
B
R
I
C
S
)
c
ount
r
ies
unde
r
s
c
or
e
the
ne
e
d
f
or
a
na
lyt
ica
l
a
pp
r
oa
c
he
s
that
e
xtend
be
yond
t
r
a
dit
ional
models
.
T
his
a
s
s
e
r
ti
on
s
ugge
s
ts
that
c
lus
ter
ing
methodologi
e
s
c
a
n
r
e
ve
a
l
de
e
pe
r
ins
ight
s
int
o
the
e
c
onomi
c
li
nka
ge
s
a
nd
pa
tt
e
r
ns
of
t
r
a
de
int
e
g
r
a
ti
o
n
a
mong
c
ount
r
ies
,
a
s
pe
c
ts
that
a
r
e
of
ten
obs
c
ur
e
d
in
c
onve
nti
ona
l
a
na
lys
e
s
.
B
y
incor
por
a
ti
ng
mac
hine
lea
r
ning
tec
hniques
a
nd
r
e
c
ognizing
the
int
r
ica
te
na
tur
e
of
tr
a
de
f
lows
,
r
e
s
e
a
r
c
he
r
s
c
a
n
be
tt
e
r
unde
r
s
tand
global
tr
a
de
dyna
mi
c
s
,
ther
e
by
e
quippi
ng
poli
c
ymake
r
s
a
nd
s
t
a
ke
holder
s
with
ins
ight
s
int
o
the
e
c
onomi
c
de
ve
lopm
e
nt
a
nd
s
us
taina
bil
it
y
im
pli
c
a
ti
ons
of
t
r
a
de
pa
tt
e
r
ns
.
Give
n
thi
s
ba
c
kdr
op,
the
pr
im
a
r
y
r
e
s
e
a
r
c
h
que
s
ti
ons
thi
s
s
tudy
s
e
e
ks
to
a
ddr
e
s
s
is
:
How
c
a
n
c
lus
ter
ing
tec
hniques
,
s
pe
c
if
ica
ll
y
a
hybr
id
f
r
a
mew
or
k
inv
olvi
ng
de
ns
it
y
-
ba
s
e
d
s
pa
ti
a
l
c
lus
t
e
r
ing
of
a
ppli
c
a
ti
ons
with
nois
e
(
DB
S
C
AN
)
,
e
lbow
,
a
nd
s
e
lf
-
or
ga
nizing
maps
(
S
OM
)
methods
,
e
lucida
t
e
dis
ti
nc
t
pa
tt
e
r
ns
a
nd
a
c
ti
ona
ble
ins
ight
s
f
r
om
global
tr
a
de
da
ta?
T
he
objec
ti
ve
s
o
f
thi
s
r
e
s
e
a
r
c
h
include
c
ompr
e
he
ns
ively
a
na
lyzing
global
tr
a
de
pa
tt
e
r
ns
,
inves
ti
ga
ti
ng
f
a
c
tor
s
that
c
ont
r
ibut
e
to
c
lus
ter
f
o
r
mations
,
s
uc
h
a
s
ge
ogr
a
phic
pr
oxim
it
y
,
indus
tr
ial
s
pe
c
ializa
ti
on,
a
nd
e
c
onomi
c
de
ve
lopm
e
nt
leve
ls
,
a
nd
p
r
ovidi
ng
poli
c
ymake
r
s
with
a
c
ti
ona
ble
ins
ight
s
r
e
ga
r
ding
potential
tr
a
de
pa
r
tner
s
hips
,
diver
s
if
ica
ti
on
s
tr
a
tegie
s
,
a
nd
s
e
c
tor
-
s
p
e
c
if
ic
c
oll
a
bor
a
ti
on
oppor
tuni
ti
e
s
.
B
y
s
ys
tema
ti
c
a
ll
y
e
mpl
oying
hybr
id
c
lus
ter
ing
meth
ods
,
thi
s
s
tudy
s
igni
f
ica
ntl
y
e
nha
nc
e
s
a
na
lyt
ica
l
pr
e
c
is
ion
a
nd
int
e
r
pr
e
ti
ve
c
lar
it
y
,
c
ont
r
ibut
ing
both
theor
e
ti
c
a
ll
y
a
nd
pr
a
c
ti
c
a
ll
y
to
the
f
ield
o
f
int
e
r
na
ti
ona
l
t
r
a
de
a
na
lys
is
.
2.
RE
L
AT
E
D
WORKS
T
he
a
na
lys
is
of
global
e
xpor
t
-
im
por
t
s
tr
a
tegie
s
thr
ough
da
ta
-
dr
iven
s
e
gmenta
ti
on
is
gr
e
a
tl
y
e
nha
nc
e
d
by
c
omput
a
ti
ona
l
a
ppr
oa
c
he
s
,
pa
r
ti
c
ular
ly
c
lus
ter
ing
tec
hniques
a
ppl
ied
to
tr
a
de
pa
tt
e
r
ns
.
C
lus
ter
ing
f
a
c
il
it
a
tes
the
identif
ica
ti
on
of
dis
ti
nc
t
t
r
a
de
c
omm
unit
ies
a
nd
pa
tt
e
r
ns
,
pr
ovidi
ng
e
s
s
e
nti
a
l
ins
ight
s
f
or
s
tr
a
tegic
de
c
is
ion
-
making
in
int
e
r
na
ti
ona
l
tr
a
de
.
A
f
ounda
ti
ona
l
e
leme
nt
in
unde
r
s
tanding
glo
ba
l
tr
a
de
dyna
mi
c
s
li
e
s
in
r
e
c
ogni
z
ing
it
s
c
ompl
e
x
ne
twor
k
s
tr
uc
tur
e
,
whic
h
unde
r
s
c
or
e
s
thes
e
s
ys
tems
'
int
e
r
de
pe
nde
nc
ies
a
nd
potential
vulner
a
bil
it
ies
.
T
h
e
a
ppli
c
a
ti
on
of
c
ompl
e
x
ne
twor
k
theor
y
ha
s
s
ubs
tantively
a
dva
nc
e
d
the
a
na
lys
is
of
tr
a
de
ne
two
r
ks
,
unc
ove
r
ing
their
s
tr
uc
tu
r
a
l
c
ha
r
a
c
ter
is
ti
c
s
a
nd
e
volut
ionar
y
dyna
mi
c
s
.
F
or
ins
tanc
e
,
C
ho
e
t
al
.
[
12]
a
r
gue
that
tr
a
dit
ional
mea
s
ur
e
s
of
tr
a
de
ope
nne
s
s
may
ove
r
l
ook
the
int
r
ica
te
de
pe
nde
nc
ies
e
mbedde
d
withi
n
tr
a
de
ne
twor
ks
,
noti
ng
that
a
f
oc
us
on
t
r
a
de
volum
e
a
lone
inade
qua
tely
r
e
f
lec
ts
a
c
ountr
y's
s
us
c
e
pti
bil
it
y
to
e
xter
na
l
s
hoc
ks
.
T
he
i
r
f
indi
ngs
r
e
ve
a
l
that
tr
a
de
o
pe
nne
s
s
,
in
is
olation,
wa
s
s
tatis
ti
c
a
ll
y
ins
igni
f
ica
nt
,
indi
c
a
ti
ng
that
a
n
a
na
lys
is
of
ne
twor
k
topol
ogy
is
c
r
uc
ial
f
or
a
c
c
ur
a
tely
a
s
s
e
s
s
ing
a
c
ountr
y’
s
r
e
s
il
ienc
e
withi
n
t
he
global
mar
ke
t
.
I
n
a
g
r
i
c
u
lt
u
r
a
l
t
r
a
de
,
t
he
e
vo
lu
ti
on
of
t
r
a
d
e
ne
two
r
ks
f
u
r
t
he
r
e
xe
mp
li
f
ies
t
he
s
e
in
te
r
de
p
e
n
de
nc
i
e
s
.
Q
ia
ng
e
t
al
.
[
13
]
f
o
r
ins
ta
nc
e
,
a
p
pl
y
c
o
mp
le
x
n
e
t
wo
r
k
t
he
o
r
y
to
e
xa
mi
ne
th
e
g
lo
ba
l
a
g
r
ic
ul
tu
r
a
l
t
r
a
d
e
n
e
t
wo
r
k,
i
de
n
ti
f
y
i
ng
p
ow
e
r
-
la
w
dis
t
r
ib
ut
io
ns
t
ha
t
hi
gh
l
ig
ht
th
e
f
un
da
me
nt
a
l
d
r
iv
e
r
s
o
f
t
r
a
de
d
yn
a
m
ics
.
S
i
m
il
a
r
l
y,
Li
e
t
al
.
[
14
]
e
x
pl
o
r
a
ti
on
o
f
the
g
lo
ba
l
r
i
c
e
t
r
a
de
n
e
tw
or
k
de
mo
ns
t
r
a
tes
h
ow
f
a
c
t
o
r
s
s
uc
h
a
s
c
e
n
t
r
a
l
it
y
a
nd
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
M
ac
hine
lear
ning
for
global
tr
ade
analys
is
:
a
hy
br
id
c
lus
ter
ing
appr
oac
h
us
ing
…
(
M
us
dali
fa
T
hamr
in)
3035
s
t
r
uc
tu
r
a
l
ga
ps
i
mp
a
c
t
f
oo
d
s
e
c
u
r
it
y
,
f
u
r
the
r
e
m
pha
s
iz
in
g
the
in
t
r
ica
te
in
te
r
c
on
ne
c
t
io
ns
t
ha
t
c
ha
r
a
c
te
r
i
z
e
f
oo
d
t
r
a
d
e
s
ys
tems
.
C
ol
le
c
t
iv
e
l
y
,
t
he
s
e
s
tu
di
e
s
i
ll
us
t
r
a
te
th
a
t
t
r
a
d
e
ne
tw
o
r
ks
o
pe
r
a
te
a
s
c
o
mp
lex
s
ys
t
e
ms
c
ha
r
a
c
t
e
r
iz
e
d
b
y
va
r
i
e
d
de
g
r
e
e
s
o
f
c
o
nne
c
t
iv
i
ty
a
nd
i
nte
r
de
pe
n
de
nc
e
r
a
t
he
r
tha
n
a
s
s
t
r
a
ig
ht
f
o
r
wa
r
d
l
in
e
a
r
c
on
ne
c
ti
ons
.
R
e
c
e
nt
global
e
ve
nt
s
,
pa
r
ti
c
ular
ly
the
C
OV
I
D
-
19
pa
nde
mi
c
,
ha
ve
f
ur
ther
c
ompl
ica
ted
the
topol
ogy
of
t
r
a
de
ne
twor
ks
,
unde
r
s
c
or
ing
thei
r
a
da
ptabili
ty
a
nd
r
e
s
il
ienc
e
to
e
xter
na
l
s
hoc
ks
.
Z
ha
o
a
nd
Hua
ng
[
15
]
c
ompar
a
ti
ve
s
tudy
o
f
the
in
ter
na
ti
ona
l
c
oa
l
t
r
a
de
ne
twor
k,
a
na
ly
z
e
d
p
r
e
-
a
nd
pos
t
-
pa
nde
mi
c
,
hi
ghli
ghts
s
igni
f
ica
nt
s
hif
ts
in
ne
twor
k
s
tr
uc
tur
e
,
s
howc
a
s
ing
the
a
da
ptabili
ty
of
tr
a
de
ne
twor
ks
unde
r
s
tr
e
s
s
.
S
uc
h
a
da
ptabili
ty,
int
r
ins
ic
to
c
ompl
e
x
s
ys
tems
,
r
e
s
ult
s
f
r
om
int
e
r
a
c
ti
ons
be
twe
e
n
node
s
(
c
ountr
ies
)
a
nd
li
nks
(
tr
a
de
r
e
lations
hips
)
,
p
r
oduc
ing
e
mer
ge
nt
be
ha
vi
or
s
that
a
r
e
not
r
e
a
dil
y
p
r
e
dicta
ble
ba
s
e
d
on
indi
vidual
c
omponents
a
lone.
T
his
c
ompl
e
xit
y
is
f
ur
ther
e
vi
de
nt
in
the
tr
a
de
o
f
s
pe
c
if
ic
c
omm
odit
ies
.
W
a
ng
e
t
al
.
[
16
]
f
or
ins
tanc
e
,
il
lus
tr
a
te
how
the
e
volut
ion
of
the
g
lobal
s
oybe
a
n
tr
a
de
ne
twor
k
is
s
ha
pe
d
by
f
a
c
to
r
s
s
uc
h
a
s
tr
a
de
volum
e
a
nd
pa
r
tner
r
e
lations
hips
,
ther
e
by
c
r
e
a
ti
ng
a
de
ns
e
we
b
of
int
e
r
de
pe
nde
nc
ies
.
S
im
il
a
r
ly,
Niu
e
t
al
.
[
17
]
c
ha
r
a
c
ter
i
z
e
the
global
c
r
ude
oil
tr
a
de
ne
twor
k
by
it
s
c
or
e
-
pe
r
ipher
y
s
tr
uc
tur
e
,
whe
r
e
domi
na
nt
c
ountr
ies
e
xe
r
t
s
ubs
tantial
in
f
luenc
e
ove
r
tr
a
de
f
lows
,
r
e
s
ult
ing
in
a
s
ymm
e
tr
ica
l
t
r
a
de
r
e
lations
hips
a
nd
de
pe
nde
nc
i
es
.
S
uc
h
im
ba
lanc
e
s
highl
igh
t
c
r
it
ica
l
v
ulner
a
bil
it
ies
withi
n
t
r
a
de
ne
twor
ks
that
c
a
n
i
mpac
t
global
tr
a
de
s
tabili
ty.
T
he
int
r
ica
te
topol
ogy
of
tr
a
de
ne
twor
ks
be
ne
f
it
s
s
igni
f
ica
ntl
y
f
r
om
a
dva
nc
e
d
a
na
lyt
ica
l
methods
,
s
uc
h
a
s
mul
ti
laye
r
ne
twor
k
a
na
lys
is
,
whic
h
Dupa
s
e
t
al
.
[
18
]
s
ugge
s
t
c
a
n
e
lucida
te
c
omm
unit
y
s
tr
uc
tur
e
s
a
nd
de
ns
e
ly
c
lus
ter
e
d
tr
a
ding
gr
oups
.
T
his
method
o
f
f
e
r
s
a
de
e
pe
r
c
ompr
e
he
ns
ion
of
the
int
e
r
c
onne
c
ti
ons
a
nd
de
pe
nde
nc
ies
unde
r
pinni
ng
global
tr
a
de
[
18]
.
B
y
mapping
thes
e
int
r
ica
te
r
e
lations
hips
,
r
e
s
e
a
r
c
he
r
s
c
a
n
be
tt
e
r
tr
a
c
e
the
e
volut
ion
of
tr
a
de
pa
tt
e
r
ns
ove
r
ti
me.
F
u
r
ther
mor
e
,
e
xter
na
l
s
hoc
ks
s
uc
h
a
s
f
ood
pr
ice
volatil
it
y
or
global
c
r
is
e
s
li
ke
the
C
OV
I
D
-
19
pa
nde
mi
c
s
ub
s
ta
nti
a
ll
y
inf
luenc
e
tr
a
de
dyna
mi
c
s
.
T
or
r
e
ggiani
e
t
al
.
[
19]
f
or
e
xa
mpl
e
,
dis
c
us
s
how
f
ood
pr
ice
s
hoc
ks
r
e
s
ha
pe
e
xpor
t
ba
r
r
ie
r
s
a
nd
im
por
t
ta
r
if
f
s
,
c
ons
e
que
ntl
y
a
lt
e
r
ing
global
tr
a
de
f
lows
.
S
im
i
lar
ly,
the
C
OV
I
D
-
19
pa
nd
e
mi
c
ha
s
e
xpos
e
d
s
upply
c
ha
in
vulner
a
bil
it
ies
,
highl
ight
ing
the
ne
c
e
s
s
it
y
of
r
e
s
il
ienc
e
in
tr
a
de
s
tr
a
tegy
de
v
e
lopm
e
nt.
As
de
mons
tr
a
ted
by
T
u
e
t
al.
[
20]
,
a
na
lyt
ica
l
tec
hniques
that
a
ll
ow
f
or
tempor
a
l
a
nd
s
pa
ti
a
l
c
lus
ter
ing
a
r
e
ins
tr
umenta
l
in
identif
ying
dyna
mi
c
c
omm
unit
ies
withi
n
tr
a
de
ne
two
r
ks
,
f
a
c
il
it
a
ti
ng
be
t
ter
a
da
ptation
to
s
uc
h
e
xter
na
l
s
hoc
ks
.
T
he
digi
tal
tr
a
ns
f
or
mation
of
the
global
e
c
on
o
my
ha
s
int
r
oduc
e
d
ne
w
c
ompl
e
xit
ies
in
tr
a
de
dyna
mi
c
s
.
T
he
r
is
e
of
digi
tal
mar
ke
ts
ha
s
e
nha
n
c
e
d
tr
a
de
e
f
f
icie
nc
y
a
nd
c
ompetit
ivene
s
s
,
ne
c
e
s
s
it
a
ti
ng
a
r
e
a
s
s
e
s
s
ment
of
tr
a
dit
ional
tr
a
de
models
[
21]
. E
-
c
omm
e
r
c
e
,
f
or
ins
tanc
e
,
r
e
duc
e
s
tr
a
ns
a
c
ti
on
c
os
ts
a
n
d
f
os
ter
s
ne
w
f
or
ms
of
mar
ke
t
e
nga
ge
ment,
the
r
e
by
r
e
s
ha
ping
the
int
e
r
na
ti
ona
l
t
r
a
de
lands
c
a
pe
[
22]
.
I
ntegr
a
ti
ng
digi
tal
tool
s
in
tr
a
de
a
na
lys
is
f
ur
ther
r
e
f
ines
s
e
gmenta
ti
on
s
tr
a
tegie
s
,
e
na
bli
ng
f
ir
ms
to
tar
ge
t
e
xpor
t
ini
ti
a
ti
v
e
s
mor
e
e
f
f
e
c
ti
ve
ly
by
leve
r
a
ging
r
e
a
l
-
ti
me
da
ta
a
nd
mar
k
e
t
int
e
ll
igenc
e
.
Applying
c
lus
ter
i
ng
tec
hniques
to
a
na
ly
z
e
global
e
xpor
t
-
im
por
t
s
tr
a
tegie
s
of
f
e
r
s
a
r
obus
t
f
r
a
mew
or
k
f
or
na
vigating
the
c
ompl
e
xit
ies
o
f
in
ter
na
ti
ona
l
tr
a
de
.
B
y
e
mpl
oying
a
dva
nc
e
d
ne
twor
k
a
na
lys
is
,
a
c
c
ounti
ng
f
o
r
the
im
pa
c
ts
of
e
xter
na
l
s
hoc
ks
,
a
nd
int
e
gr
a
ti
ng
digi
tal
tr
a
ns
f
or
mations
,
s
take
holder
s
a
r
e
pos
it
ioned
to
f
o
r
mul
a
te
mor
e
e
f
f
e
c
ti
ve
a
nd
r
e
s
il
ient
tr
a
de
s
tr
a
tegie
s
in
r
e
s
pons
e
to
a
n
e
volvi
ng
global
lands
c
a
pe
.
T
his
r
e
s
e
a
r
c
h
a
im
s
to
c
ons
tr
uc
t
a
c
ompr
e
he
ns
ive,
a
da
ptable
,
a
nd
pr
a
c
ti
c
a
l
f
r
a
mew
or
k
f
o
r
s
e
gmenting
global
tr
a
de
p
a
tt
e
r
ns
.
T
hr
ough
innovative
c
lus
ter
ing
a
nd
ne
twor
k
a
na
lys
is
,
thi
s
s
tudy
a
dva
nc
e
s
theor
e
ti
c
a
l
unde
r
s
tanding.
I
t
pr
ovides
a
c
ti
ona
ble
ins
ight
s
f
or
poli
c
ymake
r
s
a
nd
bus
ines
s
e
s
,
e
quippi
ng
th
e
m
with
da
ta
-
dr
iven
s
tr
a
tegie
s
f
or
f
os
ter
ing
r
e
s
il
ient
a
nd
a
d
a
pti
ve
tr
a
de
pr
a
c
ti
c
e
s
.
3.
M
E
T
HO
D
T
his
s
tudy
e
mpl
oys
s
e
ve
r
a
l
c
lus
ter
ing
a
lgor
it
hms
to
identif
y
pa
tt
e
r
ns
in
global
t
r
a
de
da
ta.
T
his
s
e
c
ti
on
de
s
c
r
ibes
the
r
e
s
e
a
r
c
h
a
pp
r
oa
c
h
us
e
d
to
p
r
ovide
a
n
ove
r
view
of
how
the
s
tudy
wa
s
c
ondu
c
ted
a
nd
wha
t
da
ta
we
r
e
c
oll
e
c
ted.
3.
1.
Dat
a
c
oll
e
c
t
ion
T
h
is
s
tu
dy
a
na
lyz
e
d
tr
a
de
da
ta
f
r
o
m
25
c
ou
n
tr
ies
c
ol
le
c
t
e
d
f
r
o
m
t
he
Un
i
te
d
Na
ti
ons
C
om
t
r
a
de
a
n
d
W
o
r
l
d
B
a
nk
r
e
pos
i
to
r
i
e
s
.
T
h
e
da
tas
e
t
s
pa
n
s
t
he
p
e
r
io
d
f
r
o
m
2
01
3
t
o
2
02
3
.
F
o
r
e
a
c
h
c
ou
nt
r
y
,
e
xp
or
t
a
nd
i
mp
o
r
t
v
a
l
ue
s
,
t
r
a
de
vo
lu
me
,
g
r
os
s
do
mes
ti
c
p
r
od
uc
t
(
GD
P
)
,
n
u
mbe
r
o
f
t
r
a
de
pa
r
tn
e
r
s
,
m
a
r
ke
t
d
is
tan
c
e
,
a
nd
t
a
r
i
f
f
r
a
tes
w
e
r
e
a
ve
r
a
g
e
d
ove
r
th
e
10
-
ye
a
r
pe
r
i
od
to
e
ns
u
r
e
c
on
s
is
te
nc
y
a
n
d
t
o
r
e
f
le
c
t
lo
ng
-
te
r
m
t
r
a
d
e
pa
t
te
r
ns
s
uit
a
b
le
f
or
c
lus
te
r
in
g
a
na
lys
is
.
3.
2.
Clu
s
t
e
r
in
g
t
e
c
h
n
iq
u
e
s
C
lus
ter
ing
is
a
n
uns
upe
r
vis
e
d
lea
r
ning
tec
hnique
us
e
d
to
gr
oup
da
ta
point
s
int
o
c
lus
ter
s
ba
s
e
d
on
s
im
il
a
r
it
y
or
s
pe
c
if
ic
c
r
it
e
r
ia.
Va
r
ious
c
lus
ter
ing
tec
hniques
a
r
e
s
uit
a
ble
f
or
dif
f
e
r
e
nt
types
of
da
ta
a
nd
r
e
s
e
a
r
c
h
objec
ti
ve
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
20
25
:
303
3
-
3046
3036
3.
2.
1
.
De
n
s
it
y
-
b
as
e
d
s
p
a
t
ial
c
lu
s
t
e
r
in
g
of
ap
p
li
c
at
ion
s
wit
h
n
ois
e
T
he
DB
S
C
AN
c
lus
ter
ing
a
lgor
it
hm
is
a
n
una
tt
e
nde
d
a
lgor
it
hm
ba
s
e
d
on
it
s
a
bil
it
y
to
de
f
ine
c
lus
ter
s
without
pr
e
de
f
ined
c
las
s
labe
ls
[
23]
.
T
he
DB
S
C
AN
c
lus
ter
ing
pr
ovides
a
r
obus
t
a
nd
f
lexible
a
p
pr
oa
c
h
to
s
e
gmenting
global
tr
a
de
ne
twor
ks
.
B
y
identi
f
yin
g
c
or
e
c
lus
ter
s
a
nd
out
l
ier
s
,
DB
S
C
AN
e
na
bles
a
nua
nc
e
d
unde
r
s
tanding
of
tr
a
de
c
omm
unit
ies
,
c
a
ptur
ing
ir
r
e
gular
it
ies
that
t
r
a
dit
ional
c
lus
ter
ing
methods
mi
g
ht
mi
s
s
.
T
his
he
lps
poli
c
ymake
r
s
a
nd
bus
ines
s
e
s
to
identif
y
ke
y
tr
a
de
r
e
lations
hips
,
r
e
s
il
ient
a
nd
vulner
a
ble
mar
ke
ts
,
a
nd
e
mer
ging
t
r
a
de
oppor
tuni
ti
e
s
withi
n
a
dy
na
mi
c
global
t
r
a
de
lands
c
a
pe
.
T
he
f
oll
owing
a
r
e
the
mathe
matica
l
f
or
mul
a
s
us
e
d
a
t
e
a
c
h
s
tage
of
the
D
B
S
C
AN
a
lgor
it
hm:
i)
De
f
ine
pa
r
a
mete
r
s
e
ps
(
ε)
:
M
a
xim
um
r
a
dius
of
the
ne
ighbor
hood
.
mi
n_s
a
mpl
e
s
:
M
ini
mum
number
of
point
s
r
e
qui
r
e
d
to
f
o
r
m
a
de
ns
e
r
e
gion.
ii)
F
ind
the
ne
ighbour
hood
of
a
point
us
ing
(
1
)
:
(
)
=
{
∈
∣
(
,
)
≤
}
(
1)
whe
r
e
(
)
is
the
ne
ighbor
hood
o
f
.
d
is
the
da
tas
e
t.
iii)
C
omput
ing
the
dis
tanc
e
(
,
)
us
ing
(
2)
:
(
,
)
=
√
∑
(
−
)
2
=
1
(
2)
whe
r
e
is
the
nu
mber
o
f
dim
e
ns
ions
,
a
nd
a
r
e
the
c
oor
dinate
s
of
a
nd
.
iv)
De
ter
mi
ning
the
labe
l
of
a
point
de
pe
nds
on
it
s
c
las
s
if
ica
ti
on
us
ing
(
3)
:
(
)
=
{
,
−
1
,
(
)
,
(
3)
3.
2.
2
.
E
lb
ow
m
e
t
h
od
C
lus
ter
a
na
lys
is
of
ten
f
a
c
e
s
c
ha
ll
e
ng
e
s
in
de
ter
mi
ning
the
opti
mal
number
o
f
c
lus
ter
s
.
C
r
e
a
ti
ng
too
many
c
lus
ter
s
c
a
n
r
e
s
ult
in
a
mi
nim
a
l
de
c
r
e
a
s
e
in
tot
a
l
c
lus
ter
va
r
iants
[
24]
.
T
he
s
tage
s
of
the
e
lbow
method
in
the
c
lus
ter
ing
p
r
oc
e
s
s
a
r
e
a
s
f
oll
ows
:
i)
B
uil
d
the
ini
ti
a
l
c
e
ntr
oid
a
nd
c
e
ntr
oids
r
a
ndoml
y
.
ii)
Alloca
te
a
ll
objec
ts
us
ing
E
uc
li
de
a
n
dis
tanc
e
(
4
)
,
a
s
f
oll
ows
:
(
,
)
=
√
∑
(
−
)
2
=
1
(
4)
whe
r
e
d(
i,
k
)
de
s
c
r
ibes
d
is
tanc
e
i
-
da
ta
to
c
e
ntr
oid
,
X
ij
r
e
late
d
index
j
-
d
a
ta,
a
nd
C
k
j
is
va
r
iable
f
or
c
e
nter
c
lus
ter
j
-
inde
x
.
iii)
R
e
c
a
lcula
te
c
lus
ter
membe
r
s
hip
us
ing
(
5)
:
=
∑
ℎ
=
1
;
ℎ
=
∈
(
4)
whe
r
e
m
is
the
number
o
f
da
ta
membe
r
s
,
a
nd
p
is
t
he
a
mount
of
da
ta
f
o
r
a
pa
r
ti
c
ular
c
e
ntr
o
id.
iv)
C
a
lcula
te
the
c
e
ntr
oid
unti
l
f
ini
s
he
d
:
t
he
e
lbow
method
is
one
of
the
methods
us
e
d
to
de
ter
mi
ne
the
opti
mal
a
mount
in
c
lus
ter
ing
a
na
lys
is
[
25]
.
E
va
lua
ti
ng
the
qua
li
ty
of
c
lus
ter
ing
with
e
lbow
is
c
a
r
r
ied
out
us
ing
(
6)
with
the
f
oll
owing
s
tage
s
:
=
∑
∑
|
|
=
|
|
2
−
1
(
6)
w
he
r
e
is
the
a
tt
r
ibut
e
va
lue
i
-
da
ta,
is
the
c
e
n
t
e
r
c
lus
ter
-
da
ta.
3.
2.
3
.
S
e
lf
-
or
gan
i
z
in
g
m
ap
s
T
he
S
OM
method
a
dd
r
e
s
s
e
s
the
li
mi
tation
of
tr
a
di
ti
ona
l
methods
,
whic
h
c
a
nnot
dir
e
c
tl
y
e
xplain
the
mi
ning
r
e
s
ult
s
f
or
high
-
dim
e
ns
ional
da
ta,
by
ma
int
a
ini
ng
the
r
e
lations
hip
be
twe
e
n
da
ta
tr
a
ns
a
c
ti
ons
[
26]
.
S
OM
c
lus
ter
ing
of
f
e
r
s
a
powe
r
f
ul
a
ppr
oa
c
h
to
e
xplor
ing
a
nd
vis
ua
li
z
ing
the
c
ompl
e
xit
y
of
g
lo
ba
l
tr
a
de
ne
twor
ks
[
27]
.
B
y
r
e
ve
a
li
ng
late
nt
s
tr
uc
tur
e
s
a
nd
c
lus
ter
s
withi
n
e
xpor
t
-
im
por
t
da
ta
,
S
OM
c
a
n
f
a
c
il
it
a
te
a
de
e
pe
r
unde
r
s
tanding
of
tr
a
de
int
e
r
de
pe
nde
nc
ies
,
identif
y
r
e
s
il
ient
tr
a
de
c
omm
uni
ti
e
s
,
a
nd
s
uppor
t
the
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
M
ac
hine
lear
ning
for
global
tr
ade
analys
is
:
a
hy
br
id
c
lus
ter
ing
appr
oac
h
us
ing
…
(
M
us
dali
fa
T
hamr
in)
3037
de
ve
lopm
e
nt
of
da
ta
-
dr
iven,
a
da
ptable
s
tr
a
tegie
s
t
ha
t
a
r
e
a
tt
une
d
to
the
dyna
mi
c
na
tur
e
of
global
tr
a
d
e
.
T
he
s
tage
s
of
the
S
OM
m
e
thod
a
r
e
a
s
f
oll
ows
[
28
]
:
i)
I
nit
iation
ne
ur
on
e
a
c
h
y1,
y3,
.
.
.
,
yn
,
then
de
ter
mi
n
ing
the
ini
ti
a
l
we
ight
.
ii)
F
ind
the
s
hor
tes
t
dis
tanc
e
us
ing
E
uc
li
d
e
a
n
dis
tanc
e
us
ing
(
7)
a
s
f
oll
ows
:
=
∑
(
−
(
7)
iii)
Upda
te
we
ight
W
ij
us
ing
(
8
)
a
s
f
ol
lows
:
(
)
=
(
)
+
(
−
(
)
)
(
8)
iv)
C
a
lcula
te
the
c
e
ntr
oid
us
ing
(
9
)
:
=
∑
(
(
)
)
∗
=
1
∑
(
)
(
9)
v)
C
a
lcula
te
the
va
lue
of
membe
r
s
hip
de
gr
e
e
s
us
ing
(
10)
:
=
∑
=
1
(
10)
vi)
De
ter
mi
ne
the
win
c
lus
ter
:
S
OM
c
ompr
is
e
s
ne
ur
ons
a
r
r
a
nge
d
in
a
n
o
r
de
r
ly,
low
-
dim
e
ns
ional
gr
id
[
29]
,
int
e
r
c
onne
c
ted
to
de
f
i
ne
map
topol
ogy
th
r
ough
input
da
t
a
int
o
a
n
or
de
r
ly
a
r
r
a
y
of
dim
e
ns
ional
node
s
.
3.
2.
4
.
H
y
b
r
id
d
e
n
s
i
t
y
-
b
as
e
d
s
p
a
t
ia
l
c
l
u
s
t
e
r
in
g
o
f
a
p
p
l
ic
a
t
i
o
n
s
w
it
h
n
o
is
e
,
e
l
b
o
w
,
a
n
d
s
e
l
f
-
or
ga
n
i
z
i
n
g
m
a
p
s
T
his
s
tudy
int
r
oduc
e
s
a
hybr
id
c
lus
ter
ing
method
ology
that
c
ombi
ne
s
DB
S
C
AN
,
e
lbow
,
a
nd
S
OM
methods
to
s
e
gment
global
e
xpor
t
-
im
por
t
tr
a
de
s
tr
a
tegie
s
with
pr
e
c
is
ion
a
nd
s
c
a
labili
ty.
T
he
DB
S
C
AN
e
f
f
e
c
ti
ve
ly
identif
ies
c
lus
ter
s
ba
s
e
d
on
de
ns
it
y,
making
i
t
s
uit
a
ble
f
or
de
tec
ti
ng
c
or
e
tr
a
ding
hubs
a
nd
pe
r
ipher
a
l
mar
ke
ts
.
T
he
e
lbow
method
p
r
ovides
r
obus
t
c
lus
ter
va
li
da
ti
on.
T
he
S
OM
is
pa
r
ti
c
ular
ly
a
de
pt
a
t
vis
ua
li
z
ing
high
-
dim
e
ns
ional,
non
-
li
ne
a
r
tr
a
de
d
a
ta,
of
f
e
r
ing
in
ter
pr
e
tative
e
a
s
e
f
o
r
c
ompl
e
x
da
t
a
s
e
ts
,
a
s
dis
c
us
s
e
d
c
ompr
e
he
ns
ively
by
Nif
ti
ye
v
a
nd
I
ba
d
oghlu
[
30]
.
B
y
ha
r
ne
s
s
ing
c
omput
a
ti
ona
l
tec
hniques
,
the
r
e
s
e
a
r
c
h
a
na
ly
s
e
s
lar
ge
-
s
c
a
le
tr
a
de
pa
tt
e
r
n
d
a
ta
s
e
ts
,
unve
il
ing
c
lus
ter
s
of
c
ountr
ies
with
s
im
il
a
r
tr
a
de
be
ha
vio
r
s
.
T
he
DB
S
C
AN
f
a
c
il
it
a
tes
de
ns
it
y
-
ba
s
e
d
c
lus
ter
ing
to
identif
y
ke
y
t
r
a
de
hubs
a
nd
outl
ier
s
,
while
the
e
lbow
method
de
ter
mi
ne
s
the
opti
mal
number
of
c
lus
ter
s
,
e
ns
ur
ing
r
obus
t
a
nd
da
ta
-
dr
iven
s
e
gmenta
ti
on.
T
he
S
OM
f
ur
ther
e
nha
nc
e
s
the
a
na
lys
is
by
c
a
ptur
ing
c
ompl
e
x,
non
-
li
ne
a
r
r
e
lations
hips
in
high
-
dim
e
ns
ional
tr
a
de
da
ta.
T
he
r
e
s
ult
ing
s
e
gmenta
ti
on
pr
ovides
a
c
ti
ona
ble
ins
ight
s
int
o
tr
a
de
s
tr
a
tegy
typol
ogies
,
e
mp
owe
r
ing
poli
c
ymake
r
s
a
nd
bus
ines
s
e
s
to
c
r
a
f
t
inf
o
r
med
a
n
d
c
ompetit
ive
s
tr
a
tegie
s
[
31]
.
T
he
de
tailed
pr
oc
e
s
s
f
low
of
the
pr
opos
e
d
methodology
is
de
picte
d
in
F
igu
r
e
1
.
F
igur
e
1.
P
r
opos
e
d
h
yb
r
id
DB
S
C
AN
,
e
lbow
,
a
nd
S
OM
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
20
25
:
303
3
-
3046
3038
T
he
e
xplana
ti
on
ba
s
e
d
on
the
il
lus
tr
a
ti
on
in
F
igu
r
e
1
is
a
s
f
oll
ows
:
i)
I
nput
Da
tas
e
ts
a
r
e
the
da
ta
to
be
gr
oupe
d
,
while
the
E
ps
il
on,
M
inP
ts
,
a
nd
L
e
a
r
ningR
a
te
pa
r
a
mete
r
s
a
r
e
us
e
d
f
or
it
e
r
a
ti
on
c
ontr
ol
a
nd
we
ight
upda
tes
.
ii)
I
nit
ializa
ti
on
I
nit
ialize
d
pa
r
ti
ti
on
matr
ix
to
ini
ti
a
ll
y
g
r
oup
da
ta
.
iii)
I
ter
a
ti
on
f
o
r
c
lus
ter
c
onve
r
ge
nc
e
:
‒
C
e
ntr
oid
c
lus
t
e
r
c
ounted.
‒
T
he
de
gr
e
e
of
da
ta
membe
r
s
hip
to
the
c
lus
ter
is
c
a
l
c
ulate
d.
‒
Obje
c
ti
ve
f
unc
ti
ons
a
r
e
e
va
luate
d
to
c
he
c
k
if
the
c
l
us
ter
is
c
onve
r
ge
nt.
iv)
I
ter
a
ti
on
f
o
r
we
ight
c
onve
r
ge
nc
e
:
‒
T
he
E
uc
li
de
a
n
dis
tanc
e
is
c
a
lcula
ted
f
o
r
e
a
c
h
da
ta.
‒
W
e
ight
s
a
r
e
upda
ted
with
the
pa
c
e
of
lea
r
n
ing.
‒
T
he
pr
oc
e
s
s
c
onti
nue
s
unti
l
the
we
ight
s
r
e
a
c
h
a
c
o
nve
r
ge
nt
va
lue.
v)
De
ter
mi
na
ti
on
of
opti
mal
c
lus
ter
with
e
lbow
method
:
‒
C
a
lcula
te
int
r
a
-
c
lus
ter
va
r
iation
f
or
dif
f
e
r
e
nt
numb
e
r
s
of
c
lus
ter
s
.
‒
I
de
nti
f
y
e
lbow
point
s
to
de
ter
mi
ne
the
op
ti
mal
nu
mber
of
c
lus
ter
s
.
vi)
Va
li
da
ti
on
a
nd
o
utput
:
‒
T
he
be
s
t
c
lus
ter
s
a
r
e
s
e
lec
ted
ba
s
e
d
on
maximum
pe
r
f
or
manc
e
.
‒
Da
ta
is
a
s
s
igned
to
the
be
s
t
c
lus
ter
.
‒
T
he
e
nd
r
e
s
ult
is
the
opti
mal
c
lus
ter
.
T
his
s
ys
tema
ti
c
method
is
idea
l
f
or
da
tas
e
ts
w
he
r
e
both
opti
mal
c
lus
ter
dis
tr
ibut
ion
a
nd
s
c
a
labili
ty
a
r
e
c
r
it
ica
l,
making
it
a
powe
r
f
ul
tool
f
or
da
ta
a
na
lys
is
a
nd
pa
tt
e
r
n
r
e
c
ognit
ion
tas
ks
.
4.
RE
S
UL
T
S
AN
D
DI
S
CU
S
S
I
ON
T
his
s
e
c
ti
on
de
s
c
r
ibes
how
to
us
e
hybr
id
DB
S
C
AN
,
e
lbow
a
nd
S
OM
,
whe
r
e
e
a
c
h
s
tep
is
c
or
r
e
late
d
unti
l
the
p
r
oc
e
s
s
f
unc
ti
ons
a
s
a
whole
.
M
ult
ipl
e
s
u
bs
e
c
ti
ons
f
e
a
tur
e
dis
c
us
s
ion
a
bout
thi
s
s
tudy.
4.
1
.
De
t
e
r
m
in
i
n
g
of
c
r
it
e
r
ia
we
igh
t
R
e
c
e
nt
a
dva
nc
e
s
in
DB
S
C
AN
ha
ve
led
to
the
de
ve
lopm
e
nt
of
hybr
id
a
lgor
it
hms
that
c
ombi
ne
DB
S
C
AN
with
other
tec
hniques
to
im
pr
ove
c
lus
t
e
r
ing
pe
r
f
or
manc
e
[
32]
.
T
he
r
e
f
or
e
,
thi
s
s
tudy
inv
ol
ve
s
the
r
ole
of
the
DB
S
C
AN
m
e
thod
in
de
ter
mi
ning
the
we
ight
of
the
c
oll
e
c
ted
dim
e
ns
ions
,
a
s
s
hown
in
T
a
ble
1.
B
a
s
e
d
on
T
a
ble
1,
ne
ur
on
we
ight
s
a
r
e
f
or
med
us
ing
the
s
c
a
le
f
unc
ti
on
on
DB
S
C
AN
.
As
is
known,
the
s
c
a
li
ng
c
r
it
e
r
ia
va
lue
in
DB
S
C
AN
c
a
n
be
us
e
d
a
s
i
nput
to
the
S
OM
method
to
gr
oup
da
ta
a
nd
upda
te
the
we
ight
of
ne
ur
ons
s
o
that
it
c
a
n
pr
oduc
e
r
e
p
r
e
s
e
ntative
c
lus
ter
r
e
s
ult
s
[
33]
.
T
he
r
e
s
ult
s
of
the
s
c
a
le
va
lue
f
or
m
a
ti
on
f
or
e
a
c
h
c
r
it
e
r
ion
a
r
e
s
hown
in
T
a
ble
2.
B
a
s
e
d
on
the
r
e
s
ult
s
in
T
a
ble
2
,
it
is
known
that
the
s
c
a
li
ng
f
unc
ti
on
in
the
DB
S
C
AN
a
lgo
r
it
hm
a
djus
ts
the
dis
tanc
e
be
tw
e
e
n
da
ta
in
da
tas
e
t
s
with
f
e
a
tur
e
s
with
dif
f
e
r
e
nt
s
c
a
les
.
I
t
is
given
e
xa
mpl
e
c
a
lcula
ti
on
of
the
mea
n
a
nd
s
tanda
r
d
de
viation
of
t
he
e
xpor
t
c
r
it
e
r
ia:
E
xpor
t_
v
a
lue
=
[
250
,
350
,
100
,
250
,
450
,
250
,
200
,
100,
70
,
300
,
200
,
150
,
40
,
200
,
60
,
150
,
150
,
150
,
150,
250,
50,
100,
250,
80,
300]
E
xpor
t_
m
ean
=
25
250
+
350
+
100
+
⋯
+
300
25
=
5310
25
=
212
.
4
S
tanda
r
d_
d
e
viation
=
√
(
250
−
212
.
4
)
2
+
(
350
−
212
.
4
)
2
+
⋯
+
(
300
−
212
.
4
)
2
25
=
55964
25
=
2238
.
56
=
√
2238
.
56
=
47
.
3
Af
ter
ge
tt
ing
the
s
tanda
r
d
de
viation
va
lue,
the
s
c
a
le
e
xpor
t
va
lue
in
Ar
ge
nti
na
(
250)
is
c
a
lcula
ted
a
s
f
oll
ows
:
S
c
a
le_
e
xpor
t
=
250
−
212
.
4
47
.
3
=
0
.
66
4.
2
.
Anal
y
z
e
glo
b
al
t
r
ad
e
p
at
t
e
r
n
s
h
yb
r
id
DB
S
CA
N,
e
lb
ow,
an
d
s
e
lf
or
gan
i
z
in
g
m
ap
s
T
he
e
lbow
a
nd
S
OM
hybr
id
a
ppr
oa
c
h
leve
r
a
ge
s
the
powe
r
of
both
the
e
lbow
method
f
or
opti
mal
gr
ouping
a
nd
a
s
e
lf
-
a
li
gna
ble
map
to
vis
ua
li
z
e
hig
h
-
dim
e
ns
ional
da
ta
[
28]
,
[
34]
.
T
h
is
s
tudy
uti
li
z
e
s
t
he
r
ole
of
the
e
lbow
method
in
de
ter
mi
ning
the
opti
mum
c
lus
ter
va
lue
(
K)
a
s
a
guideline
f
or
c
lus
ter
f
o
r
mat
ion.
T
he
r
e
s
ult
s
of
the
e
lbow
c
a
lcula
ti
on
invol
ving
the
da
ta
in
T
a
ble
2
a
r
e
s
hown
in
F
igur
e
2.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
M
ac
hine
lear
ning
for
global
tr
ade
analys
is
:
a
hy
br
id
c
lus
ter
ing
appr
oac
h
us
ing
…
(
M
us
dali
fa
T
hamr
in)
3039
T
a
ble
1.
Da
tas
e
t
f
or
e
a
c
h
c
r
it
e
r
ion
on
tr
a
de
pa
tt
e
r
n
s
No
C
ount
r
y
E
xpor
t
(
$ B
il
li
on)
I
mpor
t
(
$ B
il
li
on)
T
r
a
de
pa
r
tn
e
r
C
ount
G
D
P
(
$ B
il
li
on)
D
is
ta
nc
e
ma
r
ke
ts
(
K
m)
T
r
a
de
vol
ume
(
$ B
il
li
on)
T
a
r
if
f
r
a
te
(%)
1
A
r
ge
nt
in
a
250
300
80
600
7000
550
8.0
2
A
us
tr
a
li
a
350
400
90
1500
6000
750
5.5
3
B
a
ngl
a
de
s
h
100
150
50
200
5500
250
8.0
4
B
r
a
z
il
250
300
100
2000
7000
550
10
5
C
a
na
da
450
500
150
1800
4500
950
5.0
6
C
hi
le
250
200
80
300
7000
450
5.0
7
C
ol
ombi
a
200
250
70
400
5500
450
7.0
8
E
gypt
100
150
60
300
6500
250
8.5
9
E
th
io
pi
a
70
100
30
120
8000
170
11.0
10
I
ndi
a
300
400
130
3000
3500
700
7.5
11
I
ndone
s
ia
200
250
70
1100
5000
450
7.0
12
K
e
nya
150
200
60
300
5000
350
8.0
13
L
ib
ya
40
60
20
70
7000
100
11.5
14
M
a
la
ys
ia
200
250
80
400
4000
450
6.0
15
M
oz
a
mbi
que
60
90
30
100
9000
150
10.5
16
N
ig
e
r
ia
150
200
70
400
8000
350
9.0
17
P
a
ki
s
ta
n
150
200
60
300
4500
350
9
.0
18
P
e
r
u
150
200
60
250
6000
350
7.5
19
P
hi
li
ppi
ne
s
150
200
70
350
5000
350
7.0
20
S
out
h
A
f
r
ic
a
250
300
80
400
6000
550
7.5
21
S
uda
n
50
70
20
80
8500
120
12.0
22
T
a
nz
a
ni
a
100
150
50
200
6000
250
9.0
23
T
ha
il
a
nd
250
300
90
500
4500
550
6.5
24
U
ga
nda
80
120
40
150
7000
200
10.0
25
V
ie
tn
a
m
300
250
100
500
3500
550
5.5
T
a
ble
2.
R
e
s
ult
s
of
s
c
a
le
va
lue
f
or
mation
No
C
ount
r
y
E
xpor
t
I
mpor
t
T
r
a
de
pa
r
tn
e
r
c
ount
G
D
P
D
is
ta
nc
e
ma
r
ke
ts
T
r
a
de
vol
ume
T
a
r
if
f
r
a
te
1
A
r
ge
nt
in
a
0.66
0.72
0.34
-
0.02
0.69
0.70
-
0.03
2
A
us
tr
a
li
a
1.65
1.67
0.67
1.25
0.01
1.68
-
1.31
3
B
a
ngl
a
de
s
h
-
0.84
-
0.70
-
0.64
-
0.58
-
0.32
-
0.77
-
0.03
4
B
r
a
z
il
0.66
0.72
1.00
1.96
0.69
0.7
0.99
5
C
a
na
da
2.65
2.62
2.63
1.68
-
1.00
2.66
-
1.57
6
C
hi
le
0.66
-
0.22
0.34
-
0.44
0.69
0.21
-
1.57
7
C
ol
ombi
a
0.16
0.25
0.01
-
0.30
-
0.32
0.21
-
0.54
8
E
gypt
-
0.84
-
0.70
-
0.31
-
0.44
0.35
-
0.77
0.23
9
E
th
io
pi
a
-
1.13
-
1.17
-
1.30
-
0.70
1.36
-
1.17
1.51
10
I
ndi
a
1.15
1.67
1.98
3.38
-
1.67
1.43
-
0.29
11
I
ndone
s
ia
0.16
0.25
0.01
0.69
-
0.66
0.21
-
0.54
12
K
e
nya
-
0.34
-
0.22
-
0.31
-
0.44
-
0.66
-
0.28
-
0.03
13
L
ib
ya
-
1.43
-
1.55
-
1.62
-
0.77
0.69
-
1.51
1.76
14
M
a
la
ys
ia
0.16
0.25
0.34
-
0.3
-
1.33
0.21
-
1.06
15
M
oz
a
mbi
que
-
1.23
-
1.27
-
1.30
-
0.73
2.03
-
1.26
1.25
16
N
ig
e
r
ia
-
0.34
-
0.22
0.01
-
0.3
1.36
-
0.28
0.48
17
P
a
ki
s
ta
n
-
0.34
-
0.22
-
0.31
-
0.44
-
1.00
-
0.28
0.48
18
P
e
r
u
-
0.34
-
0.22
-
0.31
-
0.51
-
0.66
-
0.28
-
0.29
19
P
hi
li
ppi
ne
s
-
0.34
-
0.22
0.01
-
0.37
0.01
-
0.28
-
0.54
..
..
..
..
..
..
..
..
..
25
V
ie
tn
a
m
1.15
0.25
1.00
-
0.16
-
1.67
0.7
-
1.31
F
igur
e
2.
R
e
s
ult
of
e
lbow
f
o
r
de
ter
mi
ning
opti
mu
m
c
lus
ter
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
20
25
:
303
3
-
3046
3040
T
he
s
um
of
s
qua
r
e
d
e
r
r
or
s
(
S
S
E
)
is
a
n
i
mpor
tant
metr
ic
f
or
e
va
luating
the
pe
r
f
or
manc
e
of
S
OM
a
n
d
a
s
s
e
s
s
ing
the
qua
li
ty
of
c
lus
ter
ing
by
mea
s
ur
in
g
the
de
viation
of
da
ta
point
s
f
r
om
the
c
e
nter
of
e
a
c
h
c
lus
ter
[
35]
,
[
36]
.
B
a
s
e
d
on
the
il
lus
tr
a
ti
on
in
F
ig
ur
e
2,
i
t
is
known
that
the
opti
mum
va
lue
(
K)
is
3
with
a
n
S
S
E
va
lue
of
61
.
32.
T
he
r
e
s
ult
s
of
the
c
a
lcula
ti
on
of
the
E
a
c
h
f
e
a
tur
e
da
tas
e
t
r
e
pr
e
s
e
nt
a
ll
the
c
r
it
e
r
ia
s
howing
that
the
c
ountr
ies
that
a
r
e
c
lus
ter
-
1
mem
be
r
s
c
ons
is
t
of
Ar
ge
nti
na
,
B
r
a
z
il
,
a
nd
C
hil
e
with
c
e
ntr
oid
va
lues
=
(
0.
66,
0.
42
,
0
.
56,
0
.
50,
0.
69
,
0
.
54,
−
0.
0
3)
.
F
or
e
a
c
h
c
ount
r
y
a
s
s
igned
to
c
lus
ter
-
1
,
c
a
lc
ulate
the
s
qua
r
e
d
dis
tanc
e
to
the
c
e
ntr
oid
C
1
,
a
nd
i
t
is
f
oll
o
wing
the
e
xplana
ti
on:
Ar
ge
nti
na
(
1)
=
0.
66,
0
.
72,
0.
34
,
−
0
.
02,
0.
69
,
0
.
70,
−
0.
03.
||
x1−
c
1
||
2
=
(
0.
66−
0.
66)
2
+
(
0
.
72−
0.
42)
2
+
(
0
.
34
-
0.
56)
2
+
(
−
0.
02−
0.
50)
2
+
(
0.
69−
0
.
69)
2
+
(
0.
70−
0
.
54)
2
+
(
−
0.
03−
(
−
0.
03)
)
2
=
0
.
4344.
B
r
a
z
il
(
2)
=
0.
66,
0
.
72,
1.
00
,
1
.
96,
0.
69
,
0
.
70,
0.
99
.
||
x2
–
c
1
||
2
=
(
0.
66−
0.
66)
2
+
(
0
.
72−
0.
42)
2
+
(
1
.
00−
0.
56)
2
+
(
1
.
96−
0.
50)
2
+
(
0.
69−
0
.
69)
2
+
(
0.
70−
0
.
54)
2
+
(
0.
99−
(
−
0.
03
)
)
2
=
3.
5077
.
C
hil
e
(
3)
=
0.
66,
−
0
.
22,
0.
34
,
−
0.
44
,
0
.
69,
0.
21
,
−
1
.
57.
||
x3
–
c
1
||
2
=
(
0.
66−
0.
66)
2
+
(
−
0
.
22−
0.
42)
2
+
0.
34−
0
.
56)
2
+
(
−
0.
44
−
0.
50)
2
+
(
0.
69−
0.
69
)
2
+
0
.
21−
0.
54)
2
+
(
−
1.
57−
(
−
0.
03)
)
2
=
3.
7915.
S
S
E
(
C
1)
=
0.
4344+
3.
5077+
3.
7915=
7.
7336
.
S
S
E
(
C
2)
=
30.
45.
S
S
E
(
C
3)
=
23.
1364.
S
S
E
is
f
inally
de
te
r
mi
ne
d=
S
S
E
(
C
1)
+
S
S
E
(
C
2)
+
S
S
E
(
C
3)
.
S
S
E
=
7.
7336+
30.
45+
23.
1364=
61.
32
.
T
he
ne
xt
s
tage
de
ter
mi
ne
s
the
dis
tr
ibu
ti
on
o
f
e
a
c
h
c
ountr
y
withi
n
the
c
lus
ter
a
r
e
a
ba
s
e
d
on
the
c
e
ntr
oid
va
lue
us
ing
S
OM
.
P
r
e
vious
r
e
s
e
a
r
c
h
h
a
s
s
tate
d
that
S
OM
a
nd
c
lus
ter
ing
r
e
quir
e
th
a
t
da
ta
be
nor
malize
d
to
e
ns
ur
e
f
e
a
tu
r
e
s
c
ontr
ibut
e
e
qua
ll
y
[
37]
.
An
e
xa
mpl
e
of
the
c
a
lcula
ti
on
of
the
s
tanda
r
d
va
lue
is
given
f
or
f
e
a
tur
e
1
(
e
xpor
t
)
in
the
f
oll
owing
wa
y:
T
he
va
lues
f
or
C
1
a
c
r
os
s
a
ll
c
ountr
ies
a
r
e
:
[
0.
66,
1
.
65,
−
0.
84,
0
.
66,
2
.
65,
0.
66
,
0.
16
,
−
0.
84,
−
1.
13,
1.
15
,
0.
16
,
−
0.
34
,
−
1.
43
,
0.
16
,
−
1.
23
,
−
0.
3
4,
−
0.
34,
−
0.
34,
−
0
.
34,
0.
66
,
−
1.
33
,
−
0
.
84,
0.
66
,
−
1
.
03,
1.
15
]
.
De
ter
mi
ning
mi
n
a
nd
max
va
lues
:
X
m
i
n
=
−
1.
43
,
X
m
a
x
=
2
.
65.
T
he
n,
pluggi
ng
int
o
the
nor
maliza
ti
on
c
a
lcula
ti
on:
S
ubs
ti
tut
e
the
va
lues
X
’
=
2
.
65
−
(
−
1
.
43
)
0
.
66
−
(
−
1
.
43
)
.
R
e
s
ult
X
’
=
2
.
65
+
1
.
43
0
.
66
+
1
.
43
=
4
.
08
2
.
09
=
0.
385.
T
he
c
a
lcula
ti
on
r
e
s
ult
s
a
r
e
dis
playe
d
in
T
a
ble
3.
T
a
ble
3.
R
e
s
ult
of
no
r
maliza
ti
on
va
lue
f
or
e
ve
r
y
f
e
a
tur
e
No
C
ount
r
y
F
e
a
tu
r
e
1
F
e
a
tu
r
e
2
F
e
a
tu
r
e
3
F
e
a
tu
r
e
4
F
e
a
tu
r
e
5
F
e
a
tu
r
e
6
F
e
a
tu
r
e
7
1
A
r
ge
nt
in
a
0.385
0.460
0.419
0.231
0.600
0.427
0.464
2
A
us
tr
a
li
a
0.755
0.806
0.800
0.661
0.198
0.782
0.187
3
B
a
ngl
a
de
s
h
0.199
0.254
0.290
0.072
0.400
0.229
0.476
4
B
r
a
z
il
0.454
0.485
0.501
0.448
0.600
0.471
0.602
5
C
a
na
da
0.755
0.806
0.800
0.661
0.198
0.782
0.187
6
C
hi
le
0.491
0.455
0.473
0.155
0.405
0.474
0.219
7
C
ol
ombi
a
0.353
0.392
0.370
0.187
0.383
0.374
0.353
8
E
gypt
0.199
0.254
0.290
0.072
0.400
0.229
0.476
9
E
th
io
pi
a
0.065
0.081
0.081
0.025
0.826
0.074
0.823
..
..
..
..
..
..
..
..
..
25
V
ie
tn
a
m
0.471
0.488
0.511
0.140
0.142
0.474
0.166
T
he
n,
w
or
k
o
n
t
he
tr
a
in
in
g
p
r
o
c
e
s
s
to
le
a
r
n
t
he
da
ta
s
tr
u
c
t
ur
e
b
y
m
a
ppi
ng
s
im
il
a
r
d
a
ta
po
in
t
s
to
th
e
n
e
a
r
e
s
t
n
e
ur
o
n
s
i
n
th
e
n
e
t
wor
k
s
o
t
ha
t
c
ou
ntr
ie
s
w
it
h
s
i
mi
lar
f
e
a
tur
e
v
a
l
ue
s
a
r
e
g
r
o
up
e
d
i
n
t
h
e
s
a
me
c
lu
s
t
e
r
.
T
h
e
o
pti
ma
l
n
um
be
r
of
c
lu
s
t
e
r
s
i
s
3
in
th
e
S
O
M
a
nd
D
B
S
C
AN
,
wh
i
c
h
is
a
pp
li
e
d
w
it
h
a
3
×
3
gr
id
pr
o
du
c
e
d
in
9
n
od
e
s
.
C
o
mp
a
r
a
ti
v
e
s
tu
di
e
s
ha
ve
s
ho
w
n
th
a
t
D
B
S
C
AN
a
nd
S
O
M
c
a
n
e
f
f
e
c
ti
ve
ly
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
M
ac
hine
lear
ning
for
global
tr
ade
analys
is
:
a
hy
br
id
c
lus
ter
ing
appr
oac
h
us
ing
…
(
M
us
dali
fa
T
hamr
in)
3041
b
e
a
p
pli
e
d
i
n
c
lu
s
ter
in
g
ta
s
k
s
[
3
8]
.
T
h
e
D
B
S
C
AN
e
x
c
e
l
s
i
n
de
n
s
i
ty
-
b
a
s
e
d
c
l
u
s
t
e
r
i
ng,
pa
r
ti
c
u
l
a
r
l
y
i
n
no
is
y
e
n
vir
on
me
nt
s
,
wh
il
e
S
O
M
p
r
o
vid
e
s
po
we
r
f
u
l
vi
s
u
a
li
z
a
ti
on
c
a
p
a
bil
it
ie
s
f
or
hi
gh
-
di
me
n
s
i
on
a
l
da
t
a
.
T
h
e
r
e
s
u
lt
s
a
r
e
s
e
e
n
in
T
a
bl
e
4.
T
a
ble
4.
R
e
s
ult
of
the
c
lus
ter
ing
ba
s
e
d
on
node
va
l
ue
No
C
ount
r
y
N
ode
1
N
ode
2
N
ode
3
N
ode
4
N
ode
5
N
ode
6
N
ode
7
N
ode
8
N
ode
9
C
lu
s
te
r
1
A
r
ge
nt
in
a
0.572
0.332
0.611
0.418
0.888
0.39
0.202
0.339
0.991
2
2
A
us
tr
a
li
a
0.688
0.627
0.625
0.438
0.423
0.813
0.776
0.751
1.599
1
3
B
a
ngl
a
de
s
h
0.493
0.622
0.126
0.814
1.375
0.404
0.559
0.762
0.650
3
4
B
r
a
z
il
0.912
0.763
0.914
0.660
0.834
0.741
0.565
0.286
1.165
1
5
C
a
na
da
1.136
1.182
1.685
0.968
0.475
1.380
1.362
1.285
2.198
1
6
C
hi
le
0.560
0.362
0.670
0.556
1.055
0.492
0.527
0.732
1.110
2
7
C
ol
ombi
a
0.308
0.175
0.386
0.384
0.960
0.122
0.320
0.539
0.970
2
8
E
gypt
0.771
0.635
0.174
0.826
1.378
0.426
0.475
0.668
0.518
3
9
E
th
io
pi
a
1.269
1.103
0.659
1.279
1.801
0.903
0.849
0.965
0.040
3
10
I
ndi
a
1.035
1.100
1.424
0.845
0.455
1.169
1.175
1.007
1.907
1
11
I
ndone
s
ia
0.326
0.289
0.491
0.305
0.833
0.220
0.394
0.490
1.065
2
12
K
e
nya
0.482
0.442
0.179
0.612
1.165
0.229
0.441
0.635
0.813
2
13
L
ib
ya
1.338
1.215
0.711
1.387
1.910
0.992
0.994
1.105
0.264
3
14
M
a
la
ys
ia
0.141
0.348
0.583
0.431
0.915
0.384
0.615
0.771
1.224
3
15
M
oz
a
mbi
que
1.361
1.171
0.769
1.354
1.875
0.991
0.913
1.036
0.181
3
16
N
ig
e
r
ia
0.861
0.637
0.456
0.804
1.311
0.512
0.355
0.517
0.556
2
17
P
a
ki
s
ta
n
0.559
0.555
0.266
0.686
1.200
0.350
0.523
0.665
0.824
3
18
P
e
r
u
0.450
0.416
0.210
0.600
1.160
0.225
0.458
0.668
0.850
2
19
P
hi
li
ppi
ne
s
0.481
0.345
0.248
0.559
1.131
0.192
0.351
0.593
0.817
3
20
S
out
h A
f
r
ic
a
0.381
0.186
0.567
0.312
0.837
0.303
0.287
0.454
1.063
2
21
S
uda
n
1.478
1.311
0.866
1.482
1.991
1.111
1.047
1.140
0.231
3
22
T
a
nz
a
ni
a
0.790
0.684
0.156
0.869
1.420
0.458
0.537
0.719
0.506
3
23
T
ha
il
a
nd
0.113
0.258
0.676
0.253
0.724
0.404
0.541
0.655
1.278
2
24
U
ga
nda
1.036
0.894
0.412
1.077
1.615
0.680
0.677
0.826
0.242
3
25
V
ie
tn
a
m
0.269
0.480
0.863
0.450
0.754
0.626
0.795
0.885
1.483
3
B
a
s
e
d
on
the
r
e
s
ult
s
of
c
lus
ter
ing
us
ing
S
OM
,
the
c
lus
ter
divi
s
ion
is
de
ter
mi
ne
d
a
s
f
oll
ows
:
i)
C
lus
ter
1
(
Globa
l
T
r
a
de
L
e
a
de
r
s
)
:
thi
s
c
lus
ter
r
e
pr
e
s
e
nts
c
ountr
ies
with
s
tr
ong,
diver
s
if
ied,
a
nd
inf
luential
g
lobal
t
r
a
de
pa
tt
e
r
ns
,
o
f
ten
dr
ivi
ng
global
c
omm
e
r
c
e
.
I
t
c
omp
r
is
e
s
major
e
c
onomi
e
s
,
including
Aus
tr
a
li
a
,
B
r
a
z
il
,
C
a
na
da
,
a
nd
I
nd
ia,
a
l
l
of
whic
h
e
xhibi
t
dive
r
s
e
tr
a
de
s
tr
uc
tur
e
s
a
nd
hi
gh
ove
r
a
ll
tr
a
de
volum
e
s
.
ii)
C
lus
ter
2
(
E
mer
ging
T
r
a
de
P
owe
r
s
)
:
c
ountr
ies
in
thi
s
c
lus
ter
may
ha
ve
g
r
owing
tr
a
de
inf
luen
c
e
,
s
howing
potential
f
or
e
xpa
ns
ion
in
int
e
r
na
ti
ona
l
m
a
r
ke
ts
,
but
not
ye
t
a
t
the
s
c
a
le
of
the
global
lea
de
r
s
.
I
t
e
nc
ompas
s
e
s
Ar
ge
nti
na
,
C
hil
e
,
C
olom
bia,
I
ndone
s
ia,
Ke
nya
,
Nige
r
ia
,
P
e
r
u
,
S
outh
Af
r
ica
,
a
nd
T
ha
il
a
nd,
whic
h
ge
ne
r
a
ll
y
s
how
mi
d
-
r
a
nge
tr
a
de
int
e
ns
it
y
a
nd
g
r
e
a
ter
r
e
li
a
nc
e
on
p
r
im
a
r
y
goods
or
r
e
gi
on
-
s
pe
c
if
ic
pa
r
tner
s
.
iii)
C
lus
ter
3
(
Nic
he
E
xpor
te
r
s
)
:
thi
s
c
lus
te
r
c
ompr
i
s
e
s
c
ountr
ies
with
mor
e
s
pe
c
ialize
d
tr
a
de
pa
tt
e
r
ns
,
f
oc
us
ing
on
s
pe
c
if
ic
indus
tr
ies
or
r
e
gional
mar
ke
ts
,
with
a
s
maller
but
s
igni
f
ica
nt
pr
e
s
e
nc
e
in
both
e
xpor
ts
a
nd
im
po
r
ts
.
I
t
c
ompr
is
e
s
lowe
r
-
diver
s
if
ica
ti
on,
r
e
s
our
c
e
-
de
pe
nde
nt
e
c
onomi
e
s
,
includin
g
B
a
nglade
s
h,
E
gypt,
E
thi
opia,
L
ibya,
M
a
lays
ia,
M
oz
a
mbi
que
,
P
a
kis
tan,
P
hil
ippi
ne
s
,
S
uda
n
,
T
a
nz
a
nia,
Uga
nda
,
a
nd
Vie
tnam.
T
he
s
e
e
c
onomi
e
s
e
xhibi
t
na
r
r
owe
r
e
xpor
t
ba
s
e
s
a
nd
of
ten
r
e
ly
on
a
s
mall
num
be
r
of
c
omm
odit
ies
a
nd
pa
r
tner
s
.
T
he
s
e
gr
oupings
r
e
f
lec
t
va
r
iations
in
indus
tr
ial
c
a
pa
c
it
y,
r
e
gional
int
e
gr
a
ti
on
,
a
nd
t
r
a
de
de
pe
nde
nc
y
pr
of
il
e
s
a
c
r
os
s
the
global
tr
a
de
lands
c
a
pe
.
B
a
s
e
d
on
tr
a
de
theor
y
a
nd
powe
r
f
ul
c
omput
a
ti
ona
l
tec
hniques
,
the
c
lus
ter
ing
r
e
s
ult
s
c
a
n
pr
ovide
a
s
tr
a
tegic
r
oa
dmap
f
or
c
ount
r
ies
to
im
pr
ove
their
t
r
a
de
c
ompetit
ivene
s
s
a
nd
int
e
gr
a
te
int
o
the
global
e
c
onomy.
4.
3
.
Com
p
ar
is
on
a
n
alys
is
C
lus
ter
ing
a
na
lys
is
is
a
f
unda
menta
l
tec
hnique
in
da
ta
mi
ning
that
a
im
s
to
g
r
oup
s
im
il
a
r
da
ta
point
s
int
o
c
lus
ter
s
.
T
he
r
e
by
unc
ove
r
ing
the
inher
e
nt
s
tr
uc
tur
e
of
the
da
ta
to
unde
r
s
tand
it
s
e
f
f
e
c
ti
ve
ne
s
s
a
nd
a
ppli
c
a
ti
on
a
c
r
os
s
va
r
ious
da
tas
e
ts
.
T
he
c
ompar
a
ti
ve
vis
ua
li
z
a
ti
on
p
r
e
s
e
nted
in
F
igur
e
3
de
mons
tr
a
tes
the
dif
f
e
r
e
nc
e
s
in
c
lus
ter
ing
outcome
s
us
ing
k
-
me
a
ns
(
F
igur
e
3
(
a
)
)
,
hie
r
a
r
c
hica
l
c
lus
ter
ing
(
F
igur
e
3
(
b)
)
,
a
nd
our
pr
opos
e
d
DB
S
C
AN
a
nd
S
OM
hybr
id
a
ppr
oa
c
h
(
F
i
gur
e
3
(
c
)
).
Hybr
id
DB
S
C
AN
a
nd
S
OM
is
a
powe
r
f
ul
a
nd
f
l
e
xibl
e
c
lus
ter
ing
method
that
e
xc
e
ls
in
ha
ndli
ng
c
ompl
e
x,
non
-
li
ne
a
r
,
a
nd
ir
r
e
gular
da
ta
dis
tr
ibut
i
ons
.
I
t
is
pa
r
ti
c
ular
ly
us
e
f
ul
in
r
e
a
l
-
wor
ld
s
c
e
na
r
ios
whe
r
e
the
da
ta
c
ontains
nois
e
,
outl
ier
s
,
a
nd
va
r
ying
c
lus
ter
de
ns
it
ies
.
T
he
a
bil
it
y
to
vis
ua
li
z
e
c
lus
ter
s
us
i
ng
S
OM
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
20
25
:
303
3
-
3046
3042
make
s
DB
S
C
AN
a
nd
S
OM
a
s
upe
r
ior
c
hoice
c
ompar
e
d
to
tr
a
dit
ional
methods
,
s
uc
h
a
s
k
-
mea
ns
a
nd
k
-
medoids
,
whic
h
may
s
tr
uggle
with
thes
e
c
ha
ll
e
nge
s
.
I
n
ter
ms
o
f
c
lus
ter
ing
qua
li
ty
,
a
da
ptabili
t
y
to
da
ta
s
tr
uc
tur
e
,
a
nd
vis
ua
li
z
a
ti
on,
the
hybr
id
DB
S
C
AN
a
nd
S
OM
methods
a
r
e
the
mos
t
e
f
f
e
c
ti
ve
,
pr
ovidi
ng
mor
e
a
c
c
ur
a
te,
mea
ningf
ul,
a
nd
int
e
r
p
r
e
table
r
e
s
ult
s
f
o
r
c
ompl
e
x
da
tas
e
ts
.
(
a
)
(
b)
(
c
)
F
igur
e
3.
Vis
ua
li
z
a
ti
on
of
c
lus
ter
ing
us
ing
(
a
)
k
-
m
e
a
ns
,
(
b
)
hie
r
a
r
c
hica
l
c
lus
ter
,
a
nd
(
c
)
DB
S
C
AN
+
S
OM
4.
4.
L
im
i
t
at
ion
T
his
s
tudy
f
a
c
e
s
s
e
ve
r
a
l
li
m
it
a
ti
ons
that
ne
e
d
to
be
c
a
r
e
f
ull
y
c
ons
ider
e
d
whe
n
a
pplyi
ng
gr
ouping
tec
hniques
to
a
na
lyze
global
e
xpor
t
-
im
por
t
s
tr
a
teg
ies
.
F
ir
s
tl
y,
the
s
uc
c
e
s
s
of
the
gr
ouping
tec
hnique
is
highl
y
de
pe
nde
nt
on
the
qua
li
ty
a
nd
c
ompl
e
tene
s
s
of
the
tr
a
de
da
ta
.
T
he
r
e
f
or
e
,
the
los
s
or
incomplete
ne
s
s
of
da
ta
f
r
om
s
e
ve
r
a
l
c
ountr
ies
or
r
e
gions
c
a
n
hinder
a
c
c
ur
a
te
s
e
gmenta
ti
on
a
nd
potential
ly
p
r
oduc
e
bias
e
d
r
e
s
ult
s
.
S
e
c
ondly,
global
tr
a
de
is
inf
luenc
e
d
by
va
r
ious
c
ompl
e
x
f
a
c
tor
s
,
including
poli
ti
c
a
l
r
e
lations
,
s
upply
c
ha
in
dis
r
upti
ons
,
a
nd
tar
if
f
s
,
whic
h
may
not
be
f
ull
y
c
a
ptur
e
d
b
y
c
lus
ter
ing
methods
that
typi
c
a
ll
y
r
e
ly
on
qua
nti
tative
tr
a
de
da
ta
,
r
e
s
ult
ing
in
a
les
s
c
o
mpr
e
he
ns
ive
unde
r
s
tanding
of
global
e
xpor
t
a
n
d
im
por
t
s
tr
a
tegie
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.