T
E
L
KO
M
NIK
A
, V
ol
.
17
,
No.
6,
Dec
em
be
r
20
1
9,
p
p.3
0
8
6~
3
099
IS
S
N: 1
69
3
-
6
93
0
,
accr
ed
ited
F
irst
Gr
ad
e b
y K
em
en
r
istekdikti,
Decr
ee
No: 2
1/E/
K
P
T
/20
18
DOI:
10.12928/TE
LK
OM
N
IK
A
.v
1
7
i
6
.
11711
◼
30
86
Rec
ei
v
ed
Nov
e
mb
er
8
, 2
01
8
; Rev
i
s
e
d
J
un
e 2
8
, 2
0
1
9
;
A
c
c
ep
ted
J
ul
y
2
4
, 2
01
9
An
ad
apt
i
v
e
clu
sterin
g
an
d
clas
sificati
on
algo
ri
th
m
for
T
w
itt
er
dat
a
streami
ng
in
A
p
ache
Sp
ark
Raed
A.
Has
an*
1
,
Ro
y
ida
A.
Ibrah
em
A
lha
ya
li
2
,
Nashw
an
Dh
e
y
a
a
Z
ak
i
3
,
A
h
med
Hu
ss
i
en
A
li
4
1
Fa
c
u
l
ty
of
Al
-
Dou
r
T
e
c
h
n
i
c
a
l
i
n
s
ti
t
u
te
/
N
o
e
rth
e
rn
T
e
c
h
n
i
c
a
l
Uni
v
e
rs
i
ty
,
M
o
s
e
l
,
4
1
0
0
2
,
Ira
q
2
Dep
a
rt
m
e
n
t
of
Co
m
p
u
t
e
r
En
g
i
n
e
e
r
i
n
g
,
C
o
l
l
e
g
e
of
En
g
i
n
e
e
ri
n
g
,
Uni
v
e
r
s
i
ty
of
D
i
y
a
l
a
,
Diy
a
l
a
,
Ira
q
3
Uni
v
e
rs
i
ty
of
In
f
o
rm
a
ti
o
n
T
e
c
h
n
o
l
o
g
y
and
c
o
m
m
u
n
i
c
a
ti
o
n
s
,
Col
l
e
g
e
of
E
n
g
i
n
e
e
r
i
n
g
,
B
a
g
h
d
a
d
,
Ira
q
4
AL
Sa
l
a
m
Un
i
v
e
rs
i
ty
Co
l
l
e
g
e
Com
p
u
t
e
r
Sc
i
e
n
c
e
De
p
a
rt
m
e
n
t
Ba
g
h
d
a
d
,
I
ra
q
*C
o
rre
s
p
o
n
d
i
n
g
a
u
th
o
r
e
-
m
a
i
l
:
ra
e
d
.i
s
c
.s
a
@g
m
a
i
l
.
c
o
m
1
,
ro
y
i
d
a
.a
l
h
a
y
a
l
i
@e
n
g
i
n
e
e
ri
n
g
.u
o
d
i
y
a
l
a
.e
d
u
.i
q
2
,
n
a
s
h
w
a
n
a
l
a
n
i
@u
o
i
t
c
.e
d
u
.i
q
c
o
m
3
,
m
s
c
.a
h
m
e
d
.h
.a
l
i
@g
m
a
i
l
.c
o
m
4
Ab
strac
t
On
-
going
b
i
g
d
a
t
a
fr
o
m
s
o
c
i
a
l
n
e
two
r
k
s
s
i
te
s
a
l
i
k
e
Twi
tt
e
r
or
Fa
c
e
b
o
o
k
h
a
s
b
e
e
n
an
e
n
tr
a
n
c
i
n
g
h
o
ts
p
o
t
fo
r
i
n
v
e
s
t
i
g
a
ti
o
n
by
r
e
s
e
a
r
c
h
e
r
s
in
c
u
rr
e
n
t
d
e
c
a
d
e
s
as
a
r
e
s
u
l
t
of
v
a
ri
o
u
s
a
s
p
e
c
t
s
i
n
c
l
u
d
i
n
g
up
-
to
-
d
a
te
-
n
e
s
s
,
a
c
c
e
s
s
i
b
i
l
i
t
y
and
p
o
p
u
l
a
ri
t
y
;
h
o
wev
e
r
a
n
y
way
t
h
e
re
m
a
y
be
a
tra
d
e
o
ff
in
a
c
c
u
ra
c
y
.
M
o
re
o
v
e
r,
c
l
u
s
te
ri
n
g
of
twi
tt
e
r
d
a
t
a
has
c
a
u
g
h
t
th
e
a
tt
e
n
ti
o
n
of
re
s
e
a
r
c
h
e
r
s
.
As
s
u
c
h
,
an
a
l
g
o
r
i
th
m
wh
i
c
h
c
a
n
c
l
u
s
te
r
d
a
ta
wi
th
i
n
a
l
e
s
s
e
r
c
o
m
p
u
ta
ti
o
n
a
l
t
i
m
e
,
e
s
p
e
c
i
a
l
l
y
fo
r
d
a
ta
s
tre
a
m
i
n
g
is
n
e
e
d
e
d
.
Th
e
p
r
e
s
e
n
te
d
a
d
a
p
t
i
v
e
c
l
u
s
te
ri
n
g
a
n
d
c
l
a
s
s
i
fi
c
a
t
i
o
n
a
l
g
o
ri
t
h
m
is
u
s
e
d
fo
r
d
a
ta
s
tre
a
m
i
n
g
in
A
p
a
c
h
e
s
p
a
rk
to
o
v
e
r
c
o
m
e
th
e
e
x
i
s
ti
n
g
p
ro
b
l
e
m
s
is
p
r
o
c
e
s
s
e
d
in
tw
o
p
h
a
s
e
s
.
In
t
h
e
fi
rs
t
p
h
a
s
e
,
th
e
i
n
p
u
t
p
r
e
-
p
ro
c
e
s
s
e
d
twi
t
te
r
d
a
ta
is
v
i
a
b
l
y
c
l
u
s
te
r
e
d
u
t
i
l
i
z
i
n
g
an
Im
p
ro
v
e
d
Fu
z
z
y
C
-
m
e
a
n
s
c
l
u
s
t
e
r
i
n
g
a
n
d
th
e
p
ro
p
o
s
e
d
c
l
u
s
te
ri
n
g
is
a
d
d
i
ti
o
n
a
l
l
y
i
m
p
ro
v
e
d
by
an
Ad
a
p
t
i
v
e
Pa
rti
c
l
e
s
warm
o
p
ti
m
i
z
a
ti
o
n
(
PSO
)
a
l
g
o
ri
th
m
.
F
u
rth
e
r
t
h
e
c
l
u
s
te
re
d
d
a
ta
s
tre
a
m
i
n
g
is
a
s
s
e
s
s
e
d
u
t
i
l
i
z
i
n
g
s
p
a
r
k
e
n
g
i
n
e
.
In
th
e
s
e
c
o
n
d
p
h
a
s
e
,
th
e
i
n
p
u
t
p
r
e
-
p
ro
c
e
s
s
e
d
Hig
g
s
d
a
ta
is
c
l
a
s
s
i
fi
e
d
u
t
i
l
i
z
i
n
g
th
e
m
o
d
i
f
i
e
d
s
u
p
p
o
r
t
v
e
c
to
r
m
a
c
h
i
n
e
(M
SV
M
)
c
l
a
s
s
i
fi
e
r
wit
h
g
r
i
d
s
e
a
r
c
h
o
p
ti
m
i
z
a
ti
o
n
.
At
long
l
a
s
t
th
e
o
p
ti
m
i
z
e
d
i
n
fo
rm
a
ti
o
n
is
a
s
s
e
s
s
e
d
in
s
p
a
r
k
e
n
g
i
n
e
a
n
d
th
e
a
s
s
e
s
s
e
d
e
s
te
e
m
is
u
t
i
l
i
z
e
d
to
d
i
s
c
o
v
e
r
an
a
c
c
o
m
p
l
i
s
h
e
d
c
o
n
fu
s
i
o
n
m
a
tri
x
.
Th
e
p
ro
p
o
s
e
d
work
is
u
ti
l
i
z
i
n
g
Twi
t
te
r
d
a
t
a
s
e
t
and
Hi
g
g
s
d
a
ta
s
e
t
fo
r
th
e
d
a
ta
s
tre
a
m
i
n
g
in
Ap
a
c
h
e
S
p
a
rk
.
Th
e
c
o
m
p
u
ta
ti
o
n
a
l
e
x
a
m
i
n
a
ti
o
n
s
e
x
h
i
b
i
t
t
h
e
s
u
p
e
ri
o
ri
t
y
of
p
re
s
e
n
te
d
a
p
p
ro
a
c
h
c
o
m
p
a
r
i
n
g
w
i
th
th
e
e
x
i
s
ti
n
g
m
e
t
h
o
d
s
in
te
rm
s
of
p
re
c
i
s
i
o
n
,
r
e
c
a
l
l
,
F
-
s
c
o
re
,
c
o
n
v
e
rg
e
n
c
e
,
RO
C
c
u
r
v
e
a
n
d
a
c
c
u
ra
c
y
.
Key
w
ords
:
c
l
a
s
s
i
f
i
c
a
ti
o
n
,
c
l
u
s
t
e
ri
n
g
,
d
a
t
a
s
tr
e
a
m
i
n
g
,
o
p
t
i
m
i
z
a
ti
o
n
,
p
re
-
p
ro
c
e
s
s
i
n
g
Copy
righ
t
©
2
0
1
9
Uni
v
e
rsi
t
a
s
Ahm
a
d
D
a
hl
a
n.
All
righ
t
s
r
e
s
e
rve
d
.
1.
Int
r
o
d
u
ctio
n
O
v
er
r
ec
e
nt
y
e
ars
,
bu
s
i
ne
s
s
es
and
as
s
oc
i
at
i
o
ns
d
i
dn
't
ha
v
e
to
s
tore
and
p
erf
or
m
m
uc
h
tas
k
s
and
an
a
l
y
t
i
c
s
on
i
n
f
or
m
ati
on
of
the
c
l
i
en
ts
[1].
T
he
need
to
c
h
an
g
e
ev
er
y
th
i
n
g
i
nt
o
i
nf
orm
ati
on
is
q
ui
t
e
e
ng
a
ge
d
to
f
ul
f
i
l
th
e
n
ec
es
s
i
ti
es
of
the
g
en
eral
po
p
ul
ati
on
.
A
l
o
ng
th
es
e
l
i
n
es
,
bi
g
da
t
a
c
am
e
i
nto
i
m
ag
e
in
the
r
ea
l
t
i
m
e
bu
s
i
ne
s
s
e
x
a
m
i
na
ti
on
of
proc
es
s
i
ng
d
ata
.
P
r
es
en
t
l
y
,
i
nd
i
v
i
d
ua
l
s
are
c
om
m
un
i
c
ati
n
g
t
he
i
r
o
pi
ni
o
ns
thr
ou
gh
on
l
i
n
e
bl
o
gs
,
c
o
nv
ers
ati
on
f
orm
s
and
f
urther
m
ore
s
o
m
e
on
l
i
n
e
ap
pl
i
c
a
ti
o
ns
l
i
k
e
F
ac
eb
o
ok
,
T
wi
tte
r
,
and
so
on
[2,
3].
In
t
he
m
os
t
r
ec
en
t
de
c
ad
e
,
the
r
e
has
be
e
n
an
en
orm
ou
s
de
v
e
l
o
pm
en
t
in
the
ut
i
l
i
z
at
i
on
of
m
i
c
r
o
bl
og
g
i
ng
s
ta
ge
s
,
l
i
k
e
T
w
i
tte
r
[4]
whi
c
h
is
o
v
er
po
w
ere
d
by
as
t
on
i
s
h
i
n
g
s
ta
ti
s
ti
c
s
[5].
W
i
de
s
pread
i
nf
or
m
ati
on
ac
c
u
m
ul
ati
on
f
r
om
ne
w
s
s
ou
r
c
es
and
m
i
c
r
o
bl
o
gs
has
de
l
i
v
ered
hu
g
e
l
i
terar
y
da
ta
i
nf
orm
ati
on
s
tr
ea
m
s
th
at
are
tr
y
i
n
g
to
pr
oc
es
s
and
ex
am
i
ne
.
T
he
i
de
nt
i
f
i
c
ati
on
of
r
i
s
i
n
g
oc
c
as
i
on
s
f
r
om
data
s
tr
ea
m
s
,
f
or
ex
a
m
pl
e,
T
w
i
tt
er
has
go
tt
en
de
v
el
op
i
ng
c
on
s
i
de
r
at
i
on
f
r
om
an
al
y
s
ts
[6,
7].
T
w
i
tte
r
is
the
en
orm
ou
s
on
l
i
ne
s
oc
i
al
ne
t
wor
k
i
ng
webp
ag
e
tha
t
pres
um
ab
l
y
en
de
d
up
ord
i
n
ar
y
s
urf
i
ng
s
i
tes
by
a
l
arg
e
n
um
be
r
of
c
l
i
en
ts
.
T
w
i
t
ter
s
up
po
r
ts
i
ts
c
l
i
en
t
to
ex
pres
s
the
s
en
ti
m
en
ts
or
thi
nk
i
ng
wi
t
h
r
es
pe
c
t
to
c
erta
i
n
c
i
r
c
u
m
s
tan
c
es
of
r
ea
l
-
w
or
l
d
ha
p
pe
n
i
n
gs
[8]
.
T
w
i
tte
r
i
n
v
es
t
i
ga
tes
t
he
t
ho
ug
hts
by
ut
i
l
i
z
i
n
g
c
l
i
e
nt's
po
s
ts
,
bl
o
gs
,
a
nd
r
ev
i
e
w
s
to
s
up
po
r
t
n
um
er
ou
s
as
s
oc
i
at
i
on
s
w
h
i
c
h
a
r
e
hook
up
w
i
t
h
T
w
i
tt
er
f
or
en
ha
nc
i
n
g
the
c
l
i
en
t
s
en
t
i
m
en
ts
and
g
ov
ernm
en
ta
l
i
s
s
ue
s
,
and
r
e
c
o
m
m
en
de
r
f
r
a
m
ew
ork
[9
-
11].
A
pa
c
h
e
s
p
ark
is
a
q
ui
c
k
,
broad
l
y
us
ef
ul
an
d
d
i
s
tr
i
b
ute
d
proc
es
s
i
ng
pl
atf
or
m
that
uti
l
i
z
es
di
s
pe
r
s
ed
m
em
ory
ge
ne
r
a
l
i
z
ati
on
to
proc
es
s
hu
ge
v
ol
um
e
of
i
nf
or
m
ati
on
ef
f
ec
ti
v
el
y
.
A
pa
c
h
e
s
pa
r
k
is
ad
ap
ta
bl
e
an
d
v
e
r
s
ati
l
e
c
om
pu
ti
ng
f
r
am
ew
o
r
k
c
o
m
pris
es
of
e
ff
ec
ti
v
e
API
a
nd
h
i
gh
er
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NIK
A
IS
S
N: 1
69
3
-
6
93
0
◼
A
n a
da
p
ti
v
e c
l
us
teri
ng
an
d
c
l
as
s
i
fi
c
at
i
on
al
go
r
i
thm
fo
r
T
wi
tte
r
...
(
Rae
d A
. Has
a
n
)
3087
r
eq
ue
s
t
ap
pa
r
at
us
es
that
are
good
wi
th
h
ad
o
op
[12].
As
of
l
ate
,
a
na
l
y
s
i
ng
en
orm
ou
s
un
s
tr
uc
tured
i
nf
orm
ati
on
is
a
bu
s
i
ne
s
s
ne
e
d.
Cl
us
ter
a
na
l
y
s
i
s
is
o
ne
of
the
m
i
ni
n
g
i
s
s
ue
s
uti
l
i
z
e
d
f
or
i
nv
es
ti
g
ati
on
l
i
k
e
a
s
s
es
s
m
en
t
m
i
ni
ng
,
s
e
nt
i
m
en
tal
i
n
v
es
ti
ga
t
i
o
n
an
d
p
op
u
l
ari
t
y
ex
am
i
na
ti
on
[
13
].
Cur
r
e
nt
s
y
s
t
em
s
us
e
de
v
i
c
es
an
d
ad
v
an
c
es
to
proc
es
s
T
w
i
tt
er
i
nf
or
m
ati
on
whi
c
h
are
ut
i
l
i
z
i
n
g
e
v
e
nt
pr
oc
es
s
i
ng
a
nd
one
m
es
s
ag
e
at
ti
m
e
i
n
v
es
t
i
ga
t
i
o
n
[1
4].
A
s
tan
do
ut
am
on
gs
t
the
l
at
es
t
s
tud
i
es
uti
l
i
z
e
d
s
e
v
era
l
l
ea
r
n
i
n
g
f
r
a
m
ew
ork
s
[15]
s
uc
h
as
K
-
Neares
t
Ne
i
g
hb
o
ur
(
K
NN
)
,
S
up
p
ort
V
ec
tor
M
ac
hi
n
e
(
S
V
M),
Ra
nd
om
F
ores
t
(
R
F
)
,
and
Naï
v
e
B
a
y
es
(
NB
)
[
16
,
17].
T
he
RF
A
g
en
erat
es
be
tte
r
r
ec
a
l
l
,
prec
i
s
i
on
,
and
F
-
m
ea
s
ure
v
al
u
es
.
S
V
M
al
l
p
erf
or
m
ed
s
i
m
i
l
arl
y
by
a
c
hi
e
v
i
n
g
ab
o
ut
93%
ac
c
ura
c
y
in
e
v
er
y
grou
p.
In
e
v
er
y
on
e
of
th
es
e
prio
r
s
tud
i
es
i
n
v
es
ti
ga
t
i
on
s
,
c
l
as
s
i
f
i
c
ati
on
w
as
us
ed
f
or
s
pa
m
di
s
c
ov
er
y
on
T
wi
tte
r
.
T
he
an
om
al
y
i
de
nti
f
i
c
at
i
on
f
r
am
ew
ork
i
m
p
r
ov
em
en
t
is
f
or
di
s
ti
ng
ui
s
h
i
ng
s
pa
m
m
ers
on
T
w
i
tte
r
u
t
i
l
i
z
i
ng
ac
c
ou
nt
data
and
s
tr
ea
m
i
ng
t
w
e
ets
[18,
19].
T
he
m
ai
n
c
on
tr
i
bu
t
i
o
ns
c
an
be
s
tat
ed
as
:
1)
pre
-
proc
es
s
ed
uti
l
i
z
i
ng
an
Im
prov
e
d
F
u
z
z
y
C
-
m
ea
ns
c
l
us
teri
n
g
to
v
i
ab
l
y
c
l
us
ter
th
e
at
w
i
t
ter
i
nf
orm
ati
on
the
n
the
c
l
us
teri
ng
is
ad
di
t
i
on
al
l
y
i
m
prov
e
d
by
uti
l
i
z
i
ng
an
A
d
ap
t
i
v
e
P
art
i
c
l
e
s
w
arm
op
ti
m
i
z
at
i
on
(
P
S
O
)
al
go
r
i
t
hm
,
2)
p
re
-
proc
es
s
ed
i
nf
orm
ati
on
is
c
l
as
s
i
f
i
ed
ut
i
l
i
z
i
n
g
th
e
m
od
i
f
i
ed
s
up
po
r
t
v
ec
tor
m
ac
hi
ne
(
MS
V
M)
c
l
as
s
i
f
i
er
wi
th
gri
d
s
ea
r
c
h
op
ti
m
i
z
at
i
o
n.T
hi
s
arti
c
l
e
is
pres
en
te
d
in
d
i
f
f
erent
s
ec
ti
on
s
as
f
ol
l
o
w
s
:
the
r
el
a
ted
pre
v
i
ou
s
s
t
ud
i
es
to
th
e
pr
op
os
e
d
s
y
s
t
em
w
ere
r
ev
i
e
wed
in
s
e
c
ti
on
2,
w
h
i
l
e
s
ec
ti
on
3
br
i
ef
l
y
di
s
c
us
s
ed
the
s
ug
g
es
ted
ap
pro
ac
h.
In
s
ec
ti
on
4,
t
he
ex
p
erim
en
tal
r
es
ul
ts
w
ere
di
s
c
us
s
ed
w
h
i
l
e
s
ec
ti
on
5
p
r
es
en
ted
the
c
on
c
l
us
i
on
.
2.
Re
se
a
r
ch
M
eth
o
d
A
H
y
pe
r
tex
t
-
I
nd
uc
e
d
T
op
i
c
S
ea
r
c
h
(
HIT
S
)
w
as
s
ug
g
es
ted
by
L
ei
l
e
i
et
al
.
[20]
ba
s
ed
on
t
he
T
op
i
c
-
Dec
i
s
i
on
s
tr
ate
g
y
(
T
D
-
HIT
S
)
and
a
La
te
nt
D
i
r
i
c
h
l
et
A
l
l
oc
a
ti
o
n
(
LDA
)
-
ba
s
e
d
T
hree
-
S
tep
d
i
s
pl
a
y
(
T
S
-
LD
A
)
.
T
he
f
r
a
m
ew
ork
w
as
s
u
gg
es
te
d
f
or
i
nf
l
ue
nti
al
s
pre
ad
ers
de
t
ec
ti
o
n
and
i
de
nti
f
i
c
at
i
on
in
s
oc
i
a
l
m
ed
i
a
data
s
tr
e
am
s
.
T
h
e
prop
os
ed
T
DHIT
S
c
an
ea
s
i
l
y
i
de
nti
f
y
the
nu
m
be
r
of
the
m
es
as
di
ff
erent
r
el
ate
d
po
s
ts
in
a
h
u
ge
nu
m
be
r
of
po
s
ts
.
TS
-
LDA
c
an
i
d
en
t
i
f
y
po
w
erf
ul
prop
ag
at
ors
of
tr
en
d
i
ng
e
v
en
t
b
as
ed
on
th
e
c
l
i
en
t
da
ta
and
th
e
p
os
t.
On
a
T
w
i
tte
r
da
tas
et
,
th
e
r
es
u
l
ts
s
ho
wed
the
ef
f
i
c
i
en
c
y
of
th
e
s
ug
ge
s
ted
m
eth
od
s
in
e
v
e
nts
r
ec
og
n
i
ti
on
and
in
di
s
ti
ng
u
i
s
hi
ng
po
w
erf
ul
ev
e
nt
prop
ag
ato
r
s
.
S
ha
ng
s
on
g
L
i
a
ng
et
a
l
.
[
21
]
prop
os
ed
a
w
ork
f
or
ha
nd
l
i
n
g
t
he
i
s
s
ue
of
c
l
i
en
t
c
l
us
t
erin
g
wi
th
r
eg
ar
ds
to
the
i
r
d
i
s
tr
i
bu
t
ed
s
h
ort
t
ex
t
s
tr
ea
m
s
.
To
ac
qu
i
r
e
b
ett
er
c
l
i
en
t
c
l
us
ter
i
ng
pe
r
f
or
m
an
c
e,
the
y
pro
po
s
ed
a
t
w
o
-
us
er
c
oo
p
erat
i
v
e
i
nte
r
es
t
f
ol
l
o
wi
ng
m
od
el
s
that
go
f
or
f
ol
l
o
wi
ng
c
h
an
g
es
of
ev
er
y
c
l
i
en
t's
d
y
n
am
i
c
po
i
nt
di
s
s
em
i
na
ti
on
as
a
te
am
w
i
th
t
he
i
r
f
ol
l
o
w
ers
’
d
y
n
am
i
c
s
ub
j
ec
t
di
s
pe
r
s
i
on
s
,
ba
s
ed
bo
th
wi
th
r
es
p
ec
t
to
the
c
o
nte
n
t
of
c
urr
en
t
s
ho
r
t
m
es
s
ag
es
and
the
r
ec
en
t
l
y
e
v
a
l
u
ate
d
c
on
v
e
y
a
nc
es
.
T
he
y
al
s
o
s
ug
ge
s
t
ed
2
c
o
l
l
ap
s
ed
G
i
bb
s
s
am
pl
i
n
g
f
r
a
m
ew
ork
s
f
or
the
c
oo
pe
r
ate
i
nd
uc
em
en
t
of
the
d
y
n
am
i
c
ad
v
an
ta
ge
s
of
the
c
l
i
en
ts
f
or
both
s
ho
r
t
-
and
l
o
ng
-
term
c
l
us
teri
ng
r
e
l
i
an
c
e
po
i
nt
m
od
el
s
.
S
tr
ea
m
i
ng
da
ta
is
o
ne
of
t
he
c
on
s
i
de
r
at
i
on
s
ac
c
ep
ti
n
g
ho
ts
p
ots
f
or
c
on
c
ep
t
-
ev
o
l
ut
i
on
s
tud
i
es
.
At
the
po
i
nt
whe
n
an
ot
he
r
c
l
as
s
h
ap
p
en
s
in
t
he
i
nf
or
m
ati
on
s
tr
e
am
it
v
e
r
y
w
e
l
l
m
a
y
be
c
on
s
i
de
r
e
d
as
an
o
the
r
i
d
ea
thus
th
e
c
on
c
ep
t
-
ev
ol
uti
o
n.
T
ah
s
ee
n
et
a
l
.
[2
2]
hi
gh
l
i
g
hte
d
the
prob
l
em
by
c
h
arac
teri
z
i
ng
a
ne
w
c
ol
l
ab
orati
v
e
s
tr
ate
g
y
c
a
l
l
ed
“
c
l
as
s
-
ba
s
ed
’
gro
up
w
h
i
c
h
s
w
a
ps
the
c
on
v
en
t
i
o
na
l
"
c
hu
nk
-
ba
s
ed
"
m
eth
od
f
or
r
ep
eti
ti
v
e
c
l
as
s
i
d
en
t
i
f
i
c
ati
on
.
T
he
s
tud
y
di
s
c
us
s
ed
t
he
att
r
i
bu
t
e
of
t
he
2
d
i
f
f
erent
tec
h
ni
q
ue
s
in
c
l
as
s
-
ba
s
e
d
gr
ou
p
in
or
de
r
to
pro
v
i
de
the
i
r
d
eta
i
l
e
d
an
al
y
s
i
s
and
c
l
arif
i
c
ati
o
n.
T
he
y
al
s
o
prov
e
d
the
s
up
erio
r
i
t
y
of
the
"
c
l
as
s
-
ba
s
ed
"
groups
o
v
er
proc
ed
ures
by
m
ea
ns
of
ob
s
erv
ati
o
n
al
m
eth
od
ol
og
y
on
v
ari
o
us
be
nc
hm
ar
k
da
ta
ba
s
es
c
om
pris
i
ng
w
e
b
r
em
ar
k
s
as
tex
t
m
i
ni
ng
c
h
al
l
en
ge
.
Le
k
ha
et
al
.
[23]
de
v
e
l
op
ed
a
f
r
a
m
ew
ork
f
or
open
-
s
ou
r
c
e
bi
g
d
ata
c
a
l
l
ed
A
pa
c
h
e
S
pa
r
k
whi
c
h
is
a
c
l
ou
d
-
b
as
ed
f
r
am
ew
ork
that
f
oc
us
on
the
de
v
e
l
op
m
en
t
of
m
ac
hi
n
e
l
ea
r
n
i
n
g
f
r
a
m
ew
ork
w
i
th
r
es
pe
c
t
to
bi
g
da
t
a
s
tr
ea
m
i
ng
.
In
t
hi
s
f
r
a
m
ew
ork
,
the
us
er
t
weet
s
hi
s
/h
er
he
al
t
h
tr
ai
ts
and
t
he
a
pp
l
i
c
at
i
on
get
th
e
eq
u
i
v
al
en
t
pr
og
r
es
s
i
v
e
l
y
,
ex
tr
i
c
ate
s
th
e
tr
ai
ts
and
de
v
e
l
o
p
m
a
c
hi
ne
l
e
arni
ng
f
r
am
ew
o
r
k
to
an
ti
c
i
pa
te
c
l
i
en
t's
he
al
th
s
tat
us
whi
c
h
w
as
the
n
l
eg
i
ti
m
ate
l
y
i
nf
orm
ed
to
hi
m
/he
r
i
m
m
ed
i
ate
l
y
to
m
a
k
e
s
ui
tab
l
e
ac
ti
o
n.
S
en
t
hi
l
an
d
Us
ha
[2
4]
wor
k
ed
on
c
ate
go
r
i
z
i
ng
s
tr
ea
m
s
of
T
w
i
tte
r
da
ta
ba
s
ed
on
s
en
ti
m
en
t
a
na
l
y
s
i
s
us
i
ng
h
y
b
r
i
di
z
at
i
o
n.
T
he
s
tud
y
us
e
d
a
UR
L
-
ba
s
e
d
s
ec
ur
i
t
y
d
e
v
i
c
e
to
c
o
l
l
ec
t
600
m
i
l
l
i
on
o
pe
n
t
wee
ts
whi
l
e
f
ea
ture
s
e
l
ec
ti
on
was
ap
pl
i
e
d
f
or
s
en
ti
m
en
t
i
nv
es
ti
g
ati
on
.
T
he
ternar
y
c
l
as
s
i
f
i
c
ati
o
n
was
pe
r
f
or
m
ed
ba
s
ed
on
a
p
r
e
-
proc
es
s
i
ng
s
tr
ate
g
y
w
h
i
l
e
the
r
es
ul
ts
of
Evaluation Warning : The document was created with Spire.PDF for Python.
◼
IS
S
N: 16
93
-
6
93
0
T
E
L
KO
M
NIK
A
V
ol
.
17
,
No
.
6,
D
ec
em
be
r
20
19
:
30
8
6
-
3
09
9
3088
the
t
w
e
ets
s
en
t
by
t
he
u
s
ers
are
c
ol
l
ec
ted
.
T
he
n,
a
h
y
bri
di
z
at
i
o
n
ap
pr
oa
c
h
ba
s
ed
on
3
op
ti
m
i
z
at
i
o
n
m
eth
od
s
(
P
S
O
,
GA
and
DT
)
w
as
a
p
pl
i
ed
f
or
c
l
as
s
i
f
i
c
ati
o
n
ac
c
urac
y
us
i
n
g
s
en
ti
m
en
t
an
al
y
s
i
s
.
T
he
r
es
ul
ts
w
er
e
c
om
pa
r
ed
w
i
t
h
pre
v
i
ou
s
w
ork
s
,
and
th
ei
r
d
ev
el
o
pe
d
s
tr
ate
g
y
de
m
on
s
tr
ate
s
a
greate
r
t
ha
n
di
f
f
erent
c
l
as
s
i
f
i
e
r
s
an
al
y
s
i
s
.
3.
P
r
o
p
o
s
ed
M
eth
o
d
o
lo
g
y
3.1
.
P
h
as
e
1:
A
d
apt
iv
e
Cl
u
steri
n
g
f
o
r
T
w
it
t
er
Dat
a
S
t
r
ea
ms
in
A
p
ac
h
e
S
p
a
r
k
T
he
pres
en
t
ed
tec
hn
i
q
ue
c
on
s
i
s
ts
of
t
he
s
ub
s
eq
ue
n
t
s
tep
s
:
i
ni
ti
a
l
l
y
,
i
np
u
t
t
wi
tt
er
data
is
pre
-
proc
es
s
ed
us
i
ng
tok
en
i
z
ati
on
and
s
top
w
ord
r
em
ov
a
l
proc
es
s
es
.
T
he
n
th
e
pre
-
proc
es
s
ed
data
is
ef
f
ec
ti
v
el
y
c
l
us
tere
d
uti
l
i
z
i
ng
an
Im
prov
e
d
F
u
z
z
y
C
-
m
ea
ns
c
l
us
teri
ng
wi
th
A
d
ap
t
i
v
e
P
arti
c
l
e
s
w
arm
op
ti
m
i
z
at
i
o
n
(
P
S
O
)
al
go
r
i
t
hm
.
F
i
na
l
l
y
twi
tt
er
da
t
a
s
tr
ea
m
i
ng
us
i
n
g
our
propos
e
d
m
eth
od
is
ex
am
i
ne
d
in
a
p
ac
he
s
pa
r
k
en
gi
ne
.
T
he
f
l
o
w
d
i
ag
r
am
of
thi
s
prop
os
ed
t
wi
tte
r
da
t
a
s
tr
ea
m
i
ng
uti
l
i
z
i
ng
ph
as
e
1
m
eth
od
ol
o
g
y
is
g
i
v
en
in
F
i
g
ure
1.
F
i
gu
r
e
1
.
F
l
o
w
d
i
a
gram
of
p
ha
s
e
1
prop
os
ed
m
eth
od
o
l
og
y
3.1.1
.
P
r
epro
ce
s
sing
In
th
e
pro
po
s
ed
t
w
i
tte
r
d
a
ta
s
tr
ea
m
i
ng
,
at
f
i
r
s
t
the
i
np
ut
da
t
a
us
e
d
f
or
the
pr
op
os
ed
prof
i
c
i
en
t
i
nf
orm
ati
on
s
tr
ea
m
i
ng
is
tak
en
f
r
o
m
the
da
tas
et
[2
5
-
30]
.
Her
e,
is
th
e
t
w
i
tt
er
i
n
pu
t
da
tas
et
.
At
t
ha
t
p
oi
nt
the
i
n
pu
t
t
wi
tte
r
d
ata
is
prepr
oc
e
s
s
ed
uti
l
i
z
i
ng
tok
en
i
z
at
i
o
n
and
s
top
wor
d
r
em
ov
al
proc
es
s
es
w
h
i
c
h
are
uti
l
i
z
e
d
to
ex
pe
l
the
c
on
f
l
i
c
ti
ng
i
nf
orm
ati
on
or
no
i
s
y
i
nf
orm
ati
on
f
r
o
m
da
tas
et.
Inp
ut
d
ata
prepr
oc
es
s
i
ng
i
nc
or
po
r
ate
s
t
he
ac
c
om
pa
n
y
i
ng
proc
es
s
e
s
[31]
.
a.
S
y
m
bo
l
i
z
a
ti
o
n
S
y
m
bo
l
i
z
a
ti
o
n
is
the
tas
k
of
s
pl
i
tt
i
ng
t
he
i
n
pu
t
i
nf
or
m
ati
on
up
i
nt
o
pi
ec
es
,
c
al
l
e
d
t
ok
en
s
,
po
s
s
i
bl
y
in
th
e
m
ea
nti
m
e
di
s
c
ardi
n
g
c
ertai
n
c
ha
r
ac
ter
s
,
l
i
k
e
pu
nc
tua
ti
o
n.
B
as
i
c
a
l
l
y
,
tok
en
i
z
at
i
o
n
is
th
e
wa
y
t
o
w
ard
s
ep
arati
ng
t
he
g
i
v
en
tex
t
i
n
to
un
i
ts
c
al
l
ed
tok
en
s
and
it
is
ut
i
l
i
z
ed
f
or
f
urther
ha
nd
l
i
ng
.
T
he
tok
en
s
m
i
gh
t
be
w
ords
,
n
um
be
r
and
pu
nc
tua
t
i
on
s
am
pl
e
[3
2
-
35]
.
T
he
r
ea
s
on
f
or
s
y
m
bo
l
i
z
a
ti
on
is
to
ex
pe
l
a
l
l
the
pu
nc
t
ua
t
i
on
m
ar
k
s
l
i
k
e
c
om
m
as
,
f
ul
l
s
top
,
h
y
p
he
n
and
brac
k
ets
.
T
he
i
np
ut
da
t
a
af
ter
a
pp
l
y
i
n
g
the
tok
en
i
z
at
i
on
is
g
i
v
en
in
(
1)
:
̅
=
{
1
,
2
,
3
,
…
,
}
(
1
)
w
he
r
e,
̅
is
th
e
tok
en
i
z
e
d
d
a
ta
an
d
=
1
,
2
,
3
,
…
,
.
b.
S
top
w
ord
r
em
ov
a
l
A
f
ter
tok
en
i
z
i
ng
,
th
e
tok
en
i
z
ed
i
nf
orm
ati
on
(
̅
)
is
gi
v
e
n
as
the
c
on
tr
i
bu
t
i
on
f
or
s
top
wor
d
r
em
ov
i
ng
a
nd
h
ere
s
om
e
un
de
s
i
r
e
d
w
ords
ar
e
r
ej
ec
te
d
by
ut
i
l
i
z
i
ng
s
top
wor
d
e
l
i
m
i
na
ti
on
.
S
to
p
wor
ds
w
i
l
l
be
w
ords
that
ar
e
by
a
nd
l
arg
e
t
ho
u
gh
t
to
be
f
uti
l
e.
T
he
pu
r
po
s
e
f
or
thi
s
proc
ed
ure
is
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NIK
A
IS
S
N: 1
69
3
-
6
93
0
◼
A
n a
da
p
ti
v
e c
l
us
teri
ng
an
d
c
l
as
s
i
fi
c
at
i
on
al
go
r
i
thm
fo
r
T
wi
tte
r
...
(
Rae
d A
. Has
a
n
)
3089
uti
l
i
z
ed
to
av
oi
d
c
on
j
un
c
t
i
o
n,
r
el
ati
on
a
l
wor
ds
,
art
i
c
l
es
and
other
c
o
nt
i
nu
ou
s
w
ord
s
,
l
i
k
e
ad
v
erbs
,
ac
ti
on
w
ords
a
nd
ad
j
ec
ti
v
es
f
r
o
m
tex
tua
l
i
nf
or
m
ati
o
n
[36
]
.
S
om
e
of
the
as
of
ten
as
p
os
s
i
bl
e
uti
l
i
z
ed
s
to
p
wor
ds
are
"
a"
,
"
m
e"
,
"
of
"
,
'th
e',
'he',
's
he
',
'
y
ou
'.
T
he
tok
en
i
z
ed
i
nf
orm
ati
on
s
ub
s
eq
ue
nt
to
ap
pl
y
i
ng
the
s
top
w
ord
el
i
m
i
na
ti
on
is
gi
v
en
in
(
2)
:
=
{
1
,
2
,
3
,
…
,
}
(
2
)
he
r
e
,
is
the
prepr
oc
es
s
ed
s
et
of
data
af
ter
el
i
m
i
na
ti
n
g
s
top
w
ords
and
=
1
,
2
,
3
,
…
,
.
3.1.2
.
Dat
a
A
g
g
r
egat
ion
A
gg
r
e
ga
t
i
on
is
the
proc
es
s
of
s
pl
i
tti
ng
a
s
et
of
ob
j
e
c
ts
in
the
da
tas
et
i
nt
o
s
ub
s
ets
or
c
l
us
ter.
E
ac
h
s
u
bs
et
is
a
c
l
us
ter,
and
att
r
i
bu
tes
in
a
c
l
us
ter
are
s
i
m
i
l
ar
to
ea
c
h
a
no
t
he
r
.
T
he
propos
ed
m
od
i
f
i
ed
f
uz
z
y
c
l
us
teri
n
g
al
go
r
i
t
hm
(
MF
CM)
is
us
ed
f
or
eff
ec
ti
v
e
c
l
us
teri
ng
where
the
pe
r
f
orm
an
c
e
of
the
MFCM
de
p
en
ds
upon
the
u
pd
at
i
ng
t
he
m
e
m
be
r
s
hi
ps
f
un
c
ti
on
us
i
ng
s
i
gm
oi
d
f
un
c
ti
on
.
A
dd
i
ti
on
a
l
l
y
MFCM
pe
r
f
or
m
an
c
e
is
i
m
prov
ed
by
us
i
ng
s
up
po
r
t
v
al
ue
b
as
ed
ad
ap
ti
v
e
P
S
O
al
go
r
i
t
hm
.
T
he
prepr
oc
es
s
ed
data
is
op
ti
m
i
z
ed
us
i
ng
s
up
po
r
t
v
al
ue
ba
s
e
d
ad
ap
ti
v
e
PSO
al
go
r
i
t
hm
be
fore
m
od
i
f
i
ed
f
u
z
z
y
c
-
m
ea
ns
c
l
us
teri
n
g
[3
7]
.
Cl
us
teri
ng
is
the
proc
es
s
of
s
ep
arati
n
g
a
s
et
of
i
te
m
s
in
the
da
t
as
et
i
nto
s
ub
s
ets
or
c
l
us
ter.
E
v
er
y
s
ub
s
et
is
a
c
l
us
ter,
an
d
tr
ai
ts
in
a
gro
up
are
l
i
k
e
ea
c
h
a
no
t
he
r
.
T
he
propos
ed
m
od
i
f
i
ed
f
uz
z
y
c
-
m
ea
ns
c
l
us
teri
ng
a
l
go
r
i
thm
(
MFCM)
is
uti
l
i
z
ed
f
or
v
i
ab
l
e
c
l
us
teri
ng
where
the
ex
ec
ut
i
on
of
t
he
MFC
M
r
e
l
i
es
on
t
he
up
da
t
i
ng
t
he
m
em
be
r
s
hi
p
f
un
c
ti
on
s
uti
l
i
z
i
ng
s
i
gm
oi
d
f
un
c
ti
on
.
A
l
s
o
MFCM
ex
ec
u
ti
on
is
i
m
prov
e
d
by
ut
i
l
i
z
i
ng
s
up
po
r
t
v
a
l
ue
ba
s
e
d
ad
ap
t
i
v
e
PSO
[
38
]
.
a.
S
up
p
ort
v
al
ue
b
as
ed
ad
a
pti
v
e
PSO
T
he
PSO
was
d
ev
el
o
pe
d
as
a
h
eu
r
i
s
t
i
c
po
pu
l
at
i
on
-
b
as
ed
o
pti
m
i
z
a
ti
o
n
m
eth
od
whi
c
h
was
i
ns
p
i
r
ed
by
th
e
f
l
oc
k
i
ng
b
eh
a
v
i
ou
r
of
b
i
r
ds
.
T
he
P
S
O
is
pr
es
en
te
d
as
a
c
ol
l
ec
ti
o
n
of
pa
r
ti
c
l
es
w
h
i
c
h
i
nd
i
v
i
d
ua
l
l
y
r
ep
r
es
en
ts
a
po
te
nti
al
s
ol
uti
o
n
[
39
]
.
T
he
pa
r
t
i
c
l
es
p
urs
ue
a
ba
s
i
c
be
ha
v
i
or:
c
op
y
t
he
ac
c
om
pl
i
s
h
m
en
t
of
ne
i
gh
b
ou
r
i
ng
p
arti
c
l
es
an
d
i
ts
o
w
n
ac
c
om
pl
i
s
he
d
tr
i
um
ph
s
.
T
he
l
oc
at
i
on
of
a
pa
r
ti
c
l
e
is
thu
s
l
y
af
f
ec
ted
by
th
e
be
s
t
p
arti
c
l
e
in
a
n
ei
g
hb
o
urhoo
d,
′
j
us
t
as
th
e
arr
a
ng
em
en
t
f
ou
nd
′
.
P
art
i
c
l
e
po
s
i
t
i
on
is
ba
l
an
c
e
d
uti
l
i
z
i
n
g
the
ac
c
om
pa
n
y
i
ng
c
on
d
i
ti
o
n:
(
′
+
1
)
=
(
′
)
+
(
′
+
1
)
(
3
)
w
he
r
e,
t
he
v
e
l
oc
i
t
y
c
om
po
n
en
t
s
i
gn
i
f
i
es
the
s
t
ep
s
i
z
e.
T
he
v
el
oc
i
t
y
is
u
pd
ate
d
v
i
a
(
4)
:
(
′
+
1
)
=
′
(
′
)
+
1
1
{
′
−
(
′
)
}
+
2
2
{
′
−
(
′
)
}
(
4
)
w
he
r
e,
′
is
t
he
i
ne
r
t
i
a
w
e
i
gh
t,
1
and
2
are
t
he
ac
c
el
erat
i
o
n
c
oe
f
f
i
c
i
en
ts
1
,
2
∈
[
0
,
1
]
,
′
is
the
i
nd
i
v
i
d
ua
l
be
s
t
po
s
i
t
i
o
n
of
pa
r
ti
c
l
e
,
an
d
′
is
the
be
s
t
po
s
i
t
i
on
of
the
p
arti
c
l
es
.
At
tha
t
po
i
nt,
Map
t
he
l
oc
ati
o
n
of
ea
c
h
pa
r
t
i
c
l
e
i
nt
o
s
ol
uti
on
s
pa
c
e
and
ev
al
u
ate
i
ts
f
i
tne
s
s
es
tee
m
as
i
nd
i
c
ate
d
by
t
he
s
up
po
r
t
v
a
l
ue
b
as
ed
f
i
tne
s
s
f
un
c
ti
on
.
In
t
he
m
ea
nti
m
e,
up
da
te
′
and
′
po
s
i
ti
on
if
r
eq
ui
r
e
d.
T
he
s
up
p
ort
v
al
ue
is
es
ti
m
ate
d
by
uti
l
i
z
i
ng
(
5)
:
̅
=
1
∗
2
∗
.
.
.
.
.
.
1
+
2
+
.
.
.
.
.
.
(
5
)
Her
e,
̅
de
n
ote
s
t
he
s
u
pp
o
r
t
v
a
l
ue
,
1
,
2
,
…
,
s
i
gn
i
f
i
es
the
i
n
p
ut
po
pu
l
at
i
on
.
T
hi
s
up
da
ti
n
g
proc
es
s
is
pr
oc
ee
ds
un
t
i
l
a
c
r
i
teri
on
is
m
et,
u
s
ua
l
l
y
it
us
ed
f
or
f
i
nd
i
ng
op
t
i
m
u
m
s
ol
uti
on
throug
h
n
um
be
r
of
i
terat
i
o
n
s
.
T
he
ps
eu
d
o
c
o
de
of
s
up
po
r
t
v
a
l
ue
b
as
ed
ad
a
pti
v
e
PSO
al
go
r
i
t
h
m
is
gi
v
e
n
in
A
l
g
orit
hm
1
.
A
l
g
orit
hm
1:
S
up
p
ort
v
al
ue
ba
s
e a
da
pt
i
v
e
P
S
O
al
g
orit
h
m
Ste
p 1
:
I
nitia
liza
t
io
n
Set th
e
i
n
itial
s
ize
k’
= 0
Set a
p
o
p
u
latio
n
s
ize
o
f
NP
Set v
elo
citie
s
s
ize
v
j
o
f
t
h
e
in
s
ec
t
Evaluation Warning : The document was created with Spire.PDF for Python.
◼
IS
S
N: 16
93
-
6
93
0
T
E
L
KO
M
NIK
A
V
ol
.
17
,
No
.
6,
D
ec
em
be
r
20
19
:
30
8
6
-
3
09
9
3090
Set
2
:
W
h
ile
co
n
d
itio
n
n
o
t r
ea
ch
ed
Do
Fo
r
j
=
1
to
N
P
Ste
p 3
:
Ca
lcula
t
e
′
a
nd
′
E
v
alu
a
te
th
e
f
it
n
es
s
o
f
p
ar
ticle
s
u
s
in
g
(
5
)
Ste
p 4
:
Upda
t
e
po
s
it
io
n a
nd
v
elo
cit
y
C
alcu
late
th
e
p
o
s
itio
n
s
a
n
d
v
el
o
cities o
f
in
s
ec
t
u
tili
zi
n
g
(
3
)
a
n
d
(
4
)
E
n
d
Fo
r
Ste
p 5
:
I
ncre
a
s
e
t
he
g
ener
a
t
io
n c
o
un
t
k’
=
k’
+
1
E
n
d
w
h
ile
b
.
Mo
di
f
i
e
d
f
u
z
z
y
C
-
m
ea
ns
(
MFCM)
c
l
us
ter
i
ng
F
u
z
z
y
c
-
m
ea
ns
is
a
c
l
us
t
erin
g
m
eth
od
whi
c
h
pe
r
m
i
ts
th
e
s
i
t
ua
t
i
on
of
on
e
d
ata
s
et
be
l
on
g
i
ng
to
m
ore
than
on
e
c
l
us
ter
at
a
t
i
m
e.
T
he
s
u
gg
es
te
d
MFCM
c
l
us
teri
ng
prov
i
de
s
b
ett
er
c
l
us
teri
n
g
p
erf
or
m
an
c
e
c
om
pa
r
ed
to
the
c
on
v
en
ti
on
al
F
C
M
c
l
us
t
erin
g
m
eth
od
s
.
In
m
od
i
f
i
ed
f
uz
z
y
c
m
ea
ns
c
l
us
ter
i
ng
,
Let
=
{
1
,
2
,
3
,
…
,
}
be
t
he
s
et
of
data
po
i
nts
af
ter
ad
a
pti
v
e
pa
r
ti
c
l
e
s
w
arm
op
t
i
m
i
z
a
ti
on
an
d
=
{
1
,
2
,
3
,
…
,
}
be
t
he
s
e
t
of
c
en
ters
.
T
he
ps
eu
do
c
od
e
of
m
od
i
f
i
ed
f
uz
z
y
c
-
m
ea
ns
c
l
u
s
teri
ng
al
go
r
i
t
hm
is
gi
v
e
n
in
al
g
orit
hm
2,
A
l
g
orit
hm
2:
ps
eu
do
c
o
de
o
f
m
od
i
f
i
ed
f
u
z
z
y
c
-
m
ea
ns
c
l
us
teri
ng
T
he
MFCM
a
l
go
r
i
thm
al
l
ots
data
to
e
v
er
y
c
l
as
s
by
u
ti
l
i
z
i
ng
f
u
z
z
y
m
em
be
r
s
hi
ps
.
T
he
m
od
i
f
i
ed
o
bj
ec
ti
v
e
f
un
c
ti
on
f
or
pa
r
ti
t
i
on
i
n
g
th
e
i
np
u
t
da
t
as
et
i
nto
c
l
us
ters
is
d
ef
i
ne
d
as
,
=
∑
∑
(
)
=
1
=
1
‖
−
‖
2
(
6
)
i
n
(
6),
r
ep
r
es
en
ts
the
d
ata
,
is
the
ℎ
c
l
us
ter
c
e
nte
r
and
is
th
e
c
on
s
t
an
t
es
tee
m
.
W
h
ere,
s
i
gm
oi
d
f
un
c
ti
on
de
n
ote
s
the
w
e
i
g
hte
d
m
ea
n
di
s
ta
nc
e
in
c
l
us
ter
,
and
it
is
ad
ap
ted
f
or
the
ef
f
ec
ti
v
e
c
l
us
ter
i
ng
in
(
6
)
gi
v
en
by
:
=
{
∑
‖
−
‖
2
=
2
∑
=
1
}
1
2
⁄
(
7
)
T
he
f
un
c
ti
on
of
be
i
ng
m
em
be
r
s
i
gn
i
f
i
es
the
l
i
k
el
i
h
o
od
of
data
f
l
e
w
w
h
i
c
h
c
om
e
fr
o
m
s
a
m
e
c
l
us
ter.
T
he
probab
i
l
i
t
y
of
da
t
a
in
F
C
M
al
go
r
i
t
h
m
is
ba
s
ed
on
the
d
i
s
ta
nc
e
of
i
nd
i
v
i
du
a
l
M
od
i
fi
e
d
f
u
z
z
y
C
-
m
e
an
s
c
l
u
s
te
r
i
n
g
I
n
p
u
t:
i
npu
t
I
p
p
p
p
p
.
.
.
.
,
,
,
3
2
1
=
be
t
he
s
e
t
of
d
a
t
a
point
s
a
ft
e
r
a
da
pt
i
v
e
p
a
rt
i
c
l
e
s
w
a
rm
op
t
i
m
i
z
a
t
i
on
a
nd
J
q
q
q
q
q
.
.
.
.
,
,
,
3
2
1
=
be
t
he
s
e
t
of in
i
t
i
a
l
i
z
e
d
c
e
nt
e
rs
.
O
u
tp
u
t:
Cl
us
t
e
re
d dat
a
Be
gi
n
1.
I
ni
t
i
a
l
i
z
e
t
he
c
e
n
t
roi
ds
,
J
j
q
j
,
.
.
.
.
1
,
=
2.
Ca
l
c
ul
a
t
e
t
he
fuz
z
y
m
e
m
be
rs
hi
p
n
J
b
y
e
qu
a
t
i
on (
6
),
3.
A
t
J
-
s
t
e
p
:
c
a
l
c
ul
a
t
e
t
he
fuz
z
y
c
e
nt
e
rs
v
e
c
t
ors
ij
v
us
i
ng
(
8
)
4.
Comput
e
t
h
e
w
e
i
g
ht
e
d me
a
n dis
t
a
nc
e
i
us
i
ng
(7)
5.
U
pdat
e
t
he
c
l
us
t
e
r c
e
n
t
roi
ds
j
z
6.
I
f a
l
g
ori
t
hm c
onv
e
rg
e
s
t
he
n S
T
O
P
;
7.
O
t
he
rw
i
s
e
re
t
urn to s
t
e
p 2
unti
l
t
h
e
a
l
g
ori
t
hm c
onv
e
rg
e
s
;
8.
re
t
urn {Clus
t
e
r}
En
d
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NIK
A
IS
S
N: 1
69
3
-
6
93
0
◼
A
n a
da
p
ti
v
e c
l
us
teri
ng
an
d
c
l
as
s
i
fi
c
at
i
on
al
go
r
i
thm
fo
r
T
wi
tte
r
...
(
Rae
d A
. Has
a
n
)
3091
i
ns
ec
t
w
i
th
ot
he
r
te
am
in
s
am
e
c
l
us
ter.
T
he
f
un
c
ti
on
s
of
m
e
m
be
r
s
hi
p
and
c
l
us
t
er
c
en
ter
v
ec
tors
are
up
da
t
ed
by
t
he
v
e
l
oc
i
t
y
and
pa
r
ti
c
l
e
p
os
i
t
i
on
s
by
(
8)
and
(
9).
=
1
∑
(
‖
−
⁄
‖
‖
−
⁄
‖
)
2
−
1
=
1
(
8
)
the
c
l
us
ters
c
en
tr
oi
d
v
al
ue
s
are
c
om
pu
ted
by
u
ti
l
i
z
i
ng
(
9)
=
∑
.
=
1
∑
=
1
(
9
)
al
g
orit
hm
wi
l
l
c
o
nti
nu
e
r
un
ni
n
g
t
i
l
l
th
e
c
ha
ng
e
b
et
w
e
e
n
t
w
o
i
t
erati
on
s
r
e
ac
h
th
e
,
f
or
the
g
i
v
en
s
en
s
i
ti
v
i
t
y
thres
ho
l
d.
ma
x
‖
(
)
−
(
+
1
)
‖
<
(
10
)
where,
=
a
term
i
na
ti
on
c
o
nd
i
t
i
o
n
l
y
i
n
g
in
t
he
r
an
g
e
of
0
and
1,
w
h
i
l
e
=
the
i
ter
ati
o
n
s
tep
s
.
Repe
at
th
e
s
tep
s
un
t
i
l
ef
f
i
c
i
en
t
c
l
us
teri
n
g
r
ea
c
h
ed
.
3.2
.
P
h
as
e
2:
E
f
f
ec
t
iv
e
Cl
as
sific
atio
n
f
o
r
Hig
g
s
Dat
a
S
t
r
ea
ms
in
A
p
a
che
S
p
a
r
k
In
t
he
s
ec
on
d
s
tag
e,
the
Hi
gg
s
da
ta
s
tr
e
am
i
ng
is
v
i
ab
l
y
pe
r
f
orm
ed
by
pre
-
pro
c
es
s
i
ng
the
i
np
u
t
i
nf
orm
ati
on
.
T
he
n
the
pre
-
proc
es
s
ed
i
nf
orm
ati
on
is
c
l
as
s
i
f
i
ed
u
ti
l
i
z
i
n
g
the
m
od
i
f
i
ed
s
up
po
r
t
v
ec
tor
m
ac
hi
ne
(
MS
V
M)
c
l
as
s
i
f
i
er
w
i
t
h
g
r
i
d
s
ea
r
c
h
op
t
i
m
i
z
at
i
on
.
At
l
on
g
l
as
t
the
o
pti
m
i
z
e
d
i
nf
orm
ati
on
is
as
s
es
s
ed
in
s
p
ark
en
gi
n
e
the
n
t
he
as
s
es
s
ed
es
tee
m
is
uti
l
i
z
ed
to
di
s
c
ov
er
th
e
c
o
nf
us
i
on
m
atri
x
is
ac
c
om
pl
i
s
he
d.
T
he
pro
po
s
ed
s
ta
ge
2
wor
k
ut
i
l
i
z
i
ng
H
i
gg
s
da
tas
ets
f
or
the
da
t
a
s
tr
ea
m
i
ng
in
A
pa
c
he
S
p
ark
.
T
he
f
l
ow
di
ag
r
am
of
ph
as
e
2
m
eth
o
do
l
og
y
f
or
the
ef
f
ec
ti
v
e
c
l
as
s
i
f
i
c
ati
on
of
hi
gg
s
da
ta
s
tr
ea
m
s
is
gi
v
e
n
in
F
i
g
ure
2.
F
i
gu
r
e
2
.
F
l
o
w
d
i
a
gram
of
p
ha
s
e 2
prop
os
ed
m
eth
od
o
l
og
y
3.2.1
.
P
r
epro
ce
s
sing
In
the
prop
os
ed
H
i
gg
s
da
ta
s
tr
e
am
i
ng
,
f
i
r
s
t
the
i
np
ut
i
nf
or
m
ati
on
uti
l
i
z
e
d
f
or
the
pro
po
s
ed
ef
f
ec
ti
v
e
i
nf
o
r
m
ati
on
s
tr
ea
m
i
ng
i
s
tak
en
f
r
o
m
the
da
tas
et
′
=
{
1
′
,
2
′
,
3
′
,
…
,
′
}
.
Her
e,
′
i
s
the
H
i
gg
s
i
n
pu
t
d
ata
s
et.
T
he
n
t
he
i
n
pu
t
H
i
g
gs
i
nf
orm
ati
on
i
s
pre
proc
e
s
s
ed
ut
i
l
i
z
i
ng
Evaluation Warning : The document was created with Spire.PDF for Python.
◼
IS
S
N: 16
93
-
6
93
0
T
E
L
KO
M
NIK
A
V
ol
.
17
,
No
.
6,
D
ec
em
be
r
20
19
:
30
8
6
-
3
09
9
3092
tok
en
i
z
a
ti
o
n
a
nd
s
to
p
wor
d
r
em
ov
al
proc
es
s
es
w
h
i
c
h
a
r
e
uti
l
i
z
e
d
t
o
ex
pe
l
c
o
nf
l
i
c
ti
ng
i
nf
or
m
ati
on
or
no
i
s
y
i
nf
orm
ati
on
f
r
o
m
da
tas
et.
H
ere,
th
e
i
n
pu
t
da
ta
i
s
f
i
r
s
t
prepr
oc
es
s
e
d
b
y
u
ti
l
i
z
i
ng
tok
en
i
z
a
ti
o
n
proc
es
s
g
i
v
en
i
n
(
1)
a
nd
s
ub
s
e
qu
e
ntl
y
t
ok
en
i
z
e
d
da
ta
i
s
proc
es
s
ed
b
y
ut
i
l
i
z
i
n
g
s
top
wor
d rem
ov
al
proc
es
s
gi
v
e
n i
n
(
2).
3.2.2
.
Dat
a
S
t
r
ea
min
g
Cl
a
ss
if
ica
t
ion
G
r
id S
ea
r
ch B
as
ed M
o
d
if
ied S
v
m
T
he
S
V
M
as
a
b
i
n
ar
y
c
l
as
s
i
f
i
c
ati
o
n
m
eth
od
i
s
r
e
l
i
an
t
o
n
the
s
tr
uc
t
ural
r
i
s
k
m
i
ni
m
i
z
at
i
on
ap
pro
ac
h.
T
he
S
V
M
i
n
i
ti
at
es
b
y
m
ap
pi
ng
t
he
tr
a
i
ni
n
g
da
ta
i
nt
o
a
h
y
pe
r
p
l
a
ne
whi
c
h
d
i
v
i
d
es
2 c
l
as
s
es
of
i
nf
orm
ati
on
i
n t
he
f
ea
ture
s
pa
c
e
an
d
m
ax
i
m
i
z
e t
he
ed
g
e o
f
d
i
v
i
s
i
o
n a
m
on
g i
ts
el
f
an
d
tho
s
e
f
oc
us
es
l
y
i
ng
c
l
os
es
t
to
i
t.
T
hi
s
d
ec
i
s
i
on
s
urf
ac
e
w
o
ul
d
th
en
be
ab
l
e
t
o
b
e
ut
i
l
i
z
ed
as
a
r
ea
s
on
f
or
c
ate
go
r
i
z
i
ng
u
nk
no
w
n
i
nf
orm
ati
on
[39]
.
S
V
M
c
l
as
s
i
f
i
c
ati
on
i
s
i
m
prov
e
d
b
y
us
i
n
g
ne
t
w
ork
grid
s
ea
r
c
h
op
t
i
m
i
z
ati
on
.
T
he
gr
i
d
s
ea
r
c
h
i
m
prov
e
m
en
t
ad
e
qu
a
tel
y
t
un
es
th
e
S
V
M
pa
r
am
ete
r
s
f
or the
be
tte
r
a
s
s
ort
m
en
t.
In
(
11
)
,
=
t
he
i
n
pu
t
s
pa
c
e,
∈
=
i
np
u
t
v
ec
tors
,
=
{
1
,
−
1
}
=
target
s
pa
c
e,
∈
=
c
l
as
s
es
,
and
=
{
(
1
,
1
)
,
…
,
(
,
)
}
=
tr
ai
ni
n
g
s
et.
In
the
S
V
M,
the
m
os
t
ex
tr
em
e
edge
h
y
pe
r
p
l
a
ne
ex
ec
ute
s
the
pa
r
ti
t
i
o
ni
ng
of
th
e
2
b
ou
nd
arie
s
=
{
1
,
−
1
}
,
i
.
e.
th
e
h
y
p
erpl
an
e
whi
c
h
m
ax
i
m
i
z
es
th
e
c
l
os
es
t
di
s
ta
nc
e
to
the
d
ata
p
oi
nts
an
d
pro
v
i
de
s
the
op
t
i
m
u
m
po
pu
l
ar
i
z
ati
on
on
ne
w
m
od
el
s
.
H
en
c
e,
a
ne
w
p
oi
nt
c
an
be
c
ate
go
r
i
z
e
d
by
f
i
r
s
t
de
f
i
ni
ng
the
as
s
ortm
en
t
f
un
c
ti
on
(
)
:
(
)
=
s
gn
(
∑
(
,
)
+
∈
)
(
11
)
w
he
r
e,
=
th
e
s
u
pp
ort
v
ec
t
ors
,
(
,
)
=
k
ernel
f
un
c
t
i
on
,
=
wei
g
hts
,
=
nu
m
be
r
of
tr
a
i
ni
n
g
s
a
m
pl
es
,
=
of
f
s
et
pa
r
am
ete
r
.
If
(
)
=
+
1
,
is
in
t
he
po
s
i
ti
v
e
c
l
as
s
,
if
(
)
=
−
1
,
is
in
the
ne
g
ati
v
e
c
l
as
s
.
T
r
ai
n
i
n
g
S
V
M
r
eq
ui
r
es
the
s
ol
uti
o
n
of
th
e
ac
c
om
pa
n
y
i
n
g
o
pti
m
i
z
at
i
on
i
s
s
ue
ex
pres
s
ed
in
(
12
)
a
nd
(
1
3)
so
as
to
a
tta
i
n
t
he
wei
g
ht
v
ec
tor
an
d
th
e
of
f
s
et
.
min
,
,
1
2
+
′
∑
0
(
12
)
w
he
r
e
(
1
4)
is
s
ub
j
ec
t
t
o:
(
(
)
+
)
≥
1
−
,
≥
0
(
13
)
T
he
r
ea
s
on
f
or
em
pl
o
y
i
n
g
th
e
G
a
us
s
i
an
S
V
M
w
h
i
c
h
em
pl
o
y
s
p
aram
ete
r
s
′
an
d
ga
m
m
a
(
′
)
is
to
tr
an
s
f
orm
t
he
c
om
po
ne
nt
v
ec
tor
s
p
ac
e
i
nto
the
i
nc
en
s
em
en
t
of
r
em
ote
ne
s
s
s
uc
h
that
pa
r
t
i
ti
on
c
an
be
pe
r
f
or
m
ed
w
i
t
h
h
i
g
he
r
a
c
c
urac
y
.
T
he
di
v
ers
i
on
is
ac
c
om
pl
i
s
he
d
us
i
ng
t
he
k
ernel
f
un
c
ti
on
(
,
)
=
(
)
̃
(
)
,
c
ha
r
ac
teri
z
ed
f
or
th
e
G
au
s
s
i
an
S
V
M
is
(
,
)
=
(
−
‖
−
‖
2
)
,
>
0
.
T
he
c
ho
i
c
e
of
proper
l
e
ar
ni
n
g
p
aram
ete
r
s
is
a
s
i
gn
i
f
i
c
an
t
s
tep
in
ac
q
ui
r
i
ng
v
er
y
m
uc
h
tu
ne
d
s
up
p
ort
v
ec
tor
m
ac
hi
ne
s
.
F
or
th
e
m
os
t
pa
r
t,
the
s
ett
i
n
gs
of
the
s
e
pa
r
am
ete
r
s
de
pe
nd
on
a
grid
s
e
arc
h.
T
he
ps
eu
do
c
od
e
f
or
the
op
ti
m
i
z
at
i
on
of
SVM
pa
r
am
ete
r
uti
l
i
z
i
ng
G
r
i
d
s
ea
r
c
h
f
or
better
c
l
as
s
i
f
i
c
ati
on
is
gi
v
en
in
al
g
orit
hm
3.
T
he
S
V
M
i
n
i
ti
al
i
z
e
to
m
ai
n
p
aram
ete
r
s
′
an
d
ga
m
m
a
(
′
)
an
d
t
he
proc
ed
ure
of
op
ti
m
i
z
at
i
o
n
b
y
i
s
o
l
at
i
n
g
t
h
e
h
y
p
er
-
pl
a
ne
to
ge
t
i
de
nti
c
al
wa
y
of
w
ork
ou
t
th
e
i
n
f
or
m
ati
on
a
nd
the
s
e
are
th
e
pa
r
am
ete
r
of
S
V
M
c
l
as
s
i
f
i
er
f
or
the
r
eg
ul
ari
z
ati
on
T
he
pa
r
am
ete
r
′
c
ha
r
ac
teri
z
es
the
m
i
s
tak
e
of
da
ta
f
l
ew.
W
h
en
the
v
al
ue
of
′
i
nc
r
ea
s
es
the
m
i
s
tak
e
r
ate
al
s
o
i
nc
r
ea
s
es
an
d
brin
gs
d
o
w
n
t
he
n
um
be
r
of
pe
r
m
i
tte
d
po
i
nts
i
n
the
err
or
r
an
ge
.
A
s
m
al
l
er
v
a
l
u
e
o
f
′
en
c
ou
r
a
ge
s
a b
i
gg
er er
r
or
ga
p
up
o
n t
he
i
s
o
l
at
i
on
of
th
e
h
y
p
er
-
pl
an
e.
F
or
G
au
s
s
i
a
n
S
V
M,
the
′
pa
r
am
ete
r
i
s
de
term
i
ne
d
as
i
t
af
f
ec
ts
i
ts
h
y
p
er
-
l
i
ne
ad
a
pta
bi
l
i
t
y
.
T
o
r
ed
uc
e
t
he
v
a
l
ue
s
of
′
,
the
h
y
p
er
-
pl
an
e
l
i
n
e
i
s
a
l
m
os
t
l
i
n
ea
r
,
an
d
f
or
i
nc
r
e
as
i
ng
th
e
nu
m
be
r
s
,
i
t
w
ork
s
ou
t
t
o
b
e
progr
es
s
i
v
e
l
y
c
urv
ed
.
E
x
pa
n
di
ng
t
he
v
a
l
ue
of
′
to
ov
er
-
f
i
tti
ng
on
wor
k
ou
t
da
t
a.
T
hi
s
grid
s
ea
r
c
h
b
as
ed
m
od
i
f
i
ed
S
V
M
c
l
as
s
i
f
i
c
ati
on
pro
v
i
de
s
th
e e
ff
ec
ti
v
e
proc
es
s
of
da
ta
s
tr
e
am
i
ng
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NIK
A
IS
S
N: 1
69
3
-
6
93
0
◼
A
n a
da
p
ti
v
e c
l
us
teri
ng
an
d
c
l
as
s
i
fi
c
at
i
on
al
go
r
i
thm
fo
r
T
wi
tte
r
...
(
Rae
d A
. Has
a
n
)
3093
A
l
g
orit
hm
3
:
Mo
di
f
i
e
d
S
V
M
wi
th
G
r
i
d s
ea
r
c
h
op
t
i
m
i
z
ati
on
4.
Result
s
and
Dis
cussion
T
he
i
m
pl
e
m
en
tat
i
on
of
ou
r
propos
ed
da
t
a
s
tr
ea
m
i
ng
us
i
n
g
a
da
pt
i
v
e
c
l
us
teri
n
g
an
d
c
l
as
s
i
f
i
c
ati
on
is
p
erf
or
m
ed
in
the
w
ork
i
ng
s
ta
ge
of
J
av
a
ap
ac
he
s
p
ark
.
T
he
T
w
i
tt
er
da
t
as
et
a
nd
Hi
gg
s
da
t
as
et
is
uti
l
i
z
e
d
to
as
s
es
s
the
pro
po
s
ed
t
wi
tt
er
da
ta
s
tr
e
am
i
ng
.
In
or
de
r
to
i
n
v
es
ti
g
ate
the
pe
r
f
orm
an
c
e
of
the
propos
ed
da
ta
s
tr
ea
m
i
ng
is
di
s
ti
ng
u
i
s
he
d
w
i
t
h
the
ex
i
s
t
i
n
g
arti
f
i
c
i
a
l
bee
c
ol
on
y
(
A
B
C)
op
t
i
m
i
z
ati
o
n
and
G
en
eti
c
a
l
go
r
i
thm
(
G
A
)
tec
hn
i
q
ue
s
in
r
eg
ards
of
Rec
al
l
,
P
r
ec
i
s
i
o
n,
F
-
m
ea
s
ure
and
Con
v
erge
nc
e.
4.1
.
P
er
f
o
r
man
c
e
A
n
al
ys
is
of
P
r
o
p
o
se
d
Clu
ste
r
ing
T
he
s
tat
i
s
ti
c
al
m
etri
c
s
of
F
-
s
c
ore,
prec
i
s
i
o
n,
an
d
r
ec
al
l
c
an
be
ex
pres
s
ed
in
the
ter
m
s
of
T
P
,
F
P
,
F
N,
an
d
T
N
W
he
r
e,
TP
(
tr
ue
po
s
i
ti
v
e),
FP
(
f
al
s
e
po
s
i
t
i
v
e),
FN
(
f
al
s
e
n
e
ga
ti
v
e)
an
d
TN
(
tr
ue
ne
g
ati
v
e)
es
t
ee
m
s
.
T
he
pe
r
f
orm
an
c
e
of
our
propos
ed
wor
k
is
an
al
y
s
e
d
by
ut
i
l
i
z
i
n
g
the
s
tat
i
s
ti
c
a
l
m
ea
s
ures
m
en
ti
o
ne
d
in
t
hi
s
s
ec
ti
on
.
4.1
.1
.
P
r
e
cis
ion
T
he
f
r
ac
ti
on
of
da
t
a
r
ec
og
ni
z
ed
w
h
i
c
h
are
ap
propr
i
at
e
to
the
ori
gi
na
l
da
ta
is
te
r
m
ed
as
prec
i
s
i
o
n
:
=
+
(
14
)
t
he
c
om
pa
r
i
s
on
graph
of
pr
op
os
ed
da
ta
s
tr
e
am
i
ng
us
i
ng
i
m
prov
ed
f
u
z
z
y
c
-
m
ea
ns
c
l
us
teri
ng
wi
t
h
ex
i
s
ti
n
g
F
u
z
z
y
C
-
m
ea
ns
c
l
us
teri
ng
(
F
C
M)
an
d
K
-
m
ea
ns
c
l
us
teri
n
g
in
term
s
of
prec
i
s
i
on
is
ap
pe
ared
in
F
i
g
ure
3.
It
d
ep
i
c
ts
the
pr
op
os
e
d
da
t
a
s
tr
ea
m
i
ng
us
i
ng
i
m
prov
ed
f
uz
z
y
c
-
m
ea
ns
c
l
us
teri
n
g
r
es
u
l
ti
ng
wel
l
i
n
t
erm
s
of
prec
i
s
i
on
th
an
t
he
ex
i
s
ti
n
g
F
u
z
z
y
C
-
m
ea
ns
c
l
us
teri
ng
(
F
CM)
an
d
K
-
m
ea
ns
c
l
us
teri
n
g.
4.1.
2
.
Reca
l
l
Rec
al
l
as
c
ertai
ns
th
e
f
r
ac
ti
on
of
da
t
a
whi
c
h
are
a
pp
r
o
pria
t
e
to
the
q
ue
r
y
d
ata
t
ha
t
are
ef
f
ec
ti
v
el
y
r
ec
og
ni
z
ed
.
Evaluation Warning : The document was created with Spire.PDF for Python.
◼
IS
S
N: 16
93
-
6
93
0
T
E
L
KO
M
NIK
A
V
ol
.
17
,
No
.
6,
D
ec
em
be
r
20
19
:
30
8
6
-
3
09
9
3094
=
+
(
15
)
T
he
c
o
m
pa
r
i
s
on
graph
of
propos
ed
d
ata
s
tr
ea
m
i
ng
us
i
ng
i
m
prov
ed
f
u
z
z
y
c
-
m
ea
n
s
c
l
us
teri
n
g
w
i
th
ex
i
s
ti
n
g
F
u
z
z
y
C
-
m
ea
ns
c
l
us
teri
ng
(
F
CM)
an
d
K
-
m
ea
ns
c
l
us
teri
ng
in
term
s
of
r
ec
al
l
is
ap
p
ea
r
e
d
in
F
i
g
ure
4.
It
de
pi
c
ts
t
he
prop
os
ed
da
ta
s
tr
ea
m
i
ng
us
i
n
g
i
m
prov
ed
f
u
z
z
y
c
-
m
ea
ns
c
l
us
teri
n
g
(
IFC
M)
r
es
ul
t
i
n
g
wel
l
i
n
term
s
of
r
ec
al
l
tha
n
t
he
ex
i
s
t
i
ng
F
u
z
z
y
C
-
m
ea
ns
c
l
us
teri
n
g (F
CM)
a
nd
K
-
m
e
an
s
c
l
us
ter
i
ng
.
F
i
gu
r
e
3.
C
om
pa
r
i
s
on
grap
h i
n
te
r
m
s
of
prec
i
s
i
on
F
i
gu
r
e
4.
C
om
pa
r
i
s
on
grap
h i
n
te
r
m
s
of
r
ec
al
l
4.1.
3
.
F
-
S
core
T
hi
s
v
a
l
ue
d
ete
r
m
i
ne
s
th
e
ac
c
urac
y
of
a
tes
t.
T
he
be
s
t
F
-
m
ea
s
ure
v
al
ue
is
1
w
h
i
l
e
the
wor
s
t
is
0.
F
-
m
ea
s
ure
is
c
o
m
pu
ted
us
i
ng
(
1
6).
=
2
×
+
(
16
)
T
he
c
o
m
pa
r
i
s
on
graph
of
propos
e
d
d
ata
s
tr
e
am
i
ng
us
i
ng
i
m
prov
ed
f
u
z
z
y
c
-
m
ea
ns
c
l
us
teri
n
g
wi
th
ex
i
s
ti
n
g
F
u
z
z
y
C
-
m
ea
ns
c
l
us
teri
ng
(
F
CM)
and
K
-
m
ea
ns
c
l
us
teri
n
g
in
term
s
of
F
-
s
c
ore
is
ap
pe
ared
in
F
i
g
ure
5.
It
d
ep
i
c
ts
the
pro
po
s
ed
d
ata
s
tr
ea
m
i
ng
us
i
ng
i
m
prov
ed
f
u
z
z
y
c
-
m
ea
ns
c
l
us
teri
ng
r
es
ul
t
i
n
g
wel
l
i
n
term
s
o
f
F
-
s
c
ore
t
ha
n
th
e
ex
i
s
ti
ng
F
u
z
z
y
C
-
m
ea
ns
c
l
us
teri
ng
(
F
CM)
an
d
K
-
m
ea
ns
c
l
us
teri
ng
.
4.1.
4
.
Co
n
v
er
g
ence
G
r
aph
T
he
c
on
v
erg
en
c
e
graph
of
the
s
ug
ge
s
ted
P
S
O
u
s
i
ng
d
ata
s
tr
ea
m
i
ng
w
i
th
A
B
C
op
ti
m
i
z
at
i
o
n
and
GA
tec
hn
i
qu
es
is
g
i
v
en
in
F
i
g
ure
6.
In
th
e
propos
e
d
P
S
O
s
y
s
t
em
,
the
c
o
nv
erge
nc
e
oc
c
urs
b
et
w
ee
n
f
i
tn
es
s
an
d
nu
m
be
r
of
i
tera
ti
o
ns
i
s
be
tte
r
tha
n
th
e
ex
i
s
ti
ng
A
B
C
an
d GA c
o
nv
ergenc
e
.
4.1
.5
.
Co
mp
u
t
atio
n
al
T
ime
It
i
s
t
he
qu
a
nti
t
y
of
ti
m
e
tak
en
f
or
the
c
om
pl
eti
on
of
propos
e
d
t
wi
tte
r
da
t
a
s
tr
ea
m
i
ng
.
T
he
c
o
m
pu
tat
i
o
na
l
t
i
m
e
of
da
ta
s
tr
ea
m
i
ng
i
n
s
ec
on
ds
c
an
be
o
bta
i
ne
d
f
r
o
m
the
da
ta
s
tr
ea
m
s
i
z
e
i
n b
i
t
an
d
the
bi
t
r
ate
i
n
bi
t
/s
ec
as
:
=
⁄
(
17
)
w
he
r
e,
̃
be
t
he
c
om
pu
tat
i
o
na
l
ti
m
e
of
c
l
as
s
i
f
i
c
ati
on
,
be
th
e
s
i
z
e
of
the
d
ata
s
tr
ea
m
,
be
the
B
i
t
r
at
e.
T
he
pe
r
f
orm
a
nc
e
r
es
u
l
t
of
ou
r
propos
e
d
IFC
M
wi
th
ex
i
s
t
i
ng
F
CM
an
d
K
-
m
ea
ns
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NIK
A
IS
S
N: 1
69
3
-
6
93
0
◼
A
n a
da
p
ti
v
e c
l
us
teri
ng
an
d
c
l
as
s
i
fi
c
at
i
on
al
go
r
i
thm
fo
r
T
wi
tte
r
...
(
Rae
d A
. Has
a
n
)
3095
c
l
us
teri
n
g
i
n
t
erm
s
of
c
o
m
pu
tat
i
o
na
l
ti
m
e
i
s
g
i
v
en
i
n
F
i
gu
r
e
7.
It
d
ep
i
c
ts
the
p
r
op
os
ed
da
t
a
s
tr
ea
m
i
ng
us
i
n
g
i
m
prov
ed
f
u
z
z
y
c
-
m
ea
ns
c
l
us
teri
n
g
(
I
F
CM)
ac
h
i
e
v
ed
b
ett
er
c
om
pu
tat
i
o
na
l
t
i
m
e
c
o
m
pa
r
ed
to
F
C
M
a
nd
K
-
m
ea
ns
c
l
us
teri
n
g.
T
he
c
om
pa
r
i
s
on
r
es
ul
ts
r
eg
ardi
ng
of
v
ario
us
pe
r
f
or
m
an
c
e
m
ea
s
ures
ut
i
l
i
z
i
ng
a
da
p
ti
v
e c
l
us
teri
ng
i
s
de
p
i
c
ted
i
n
T
ab
l
e 1
.
F
i
gu
r
e
5.
C
om
pa
r
i
s
on
grap
h i
n
te
r
m
s
of
F
-
s
c
ore
F
i
gu
r
e
6.
C
on
v
erge
nc
e g
r
a
ph
of
prop
os
ed
P
S
O
uti
l
i
z
e
d c
l
us
ter
i
ng
w
i
t
h
ex
i
s
ti
n
g A
B
C
an
d G
A
te
c
h
ni
qu
es
F
i
gu
r
e
7
.
C
om
pa
r
i
s
on
grap
h
in
term
s
of
c
o
m
pu
tat
i
o
na
l
ti
m
e
4.2
.
A
v
er
age
Cl
as
s
if
ica
t
ion
E
r
r
o
r
P
er
ce
n
t
age
T
he
c
o
m
pa
r
i
s
on
as
s
es
s
m
en
t
of
the
c
l
as
s
i
f
i
c
ati
o
n
err
o
r
pe
r
c
en
ta
ge
is
gi
v
e
n
in
T
ab
l
e
2.
T
he
propos
ed
m
od
i
f
i
ed
s
up
po
r
t
v
ec
tor
m
ac
hi
ne
(
M
S
V
M)
c
l
as
s
i
f
i
c
ati
o
n
err
or
pe
r
c
en
ta
ge
is
s
i
gn
i
f
i
c
an
tl
y
l
es
s
er
tha
n
th
e
ex
i
s
ti
n
g
S
V
M
an
d
A
nti
-
B
a
y
es
Mu
l
ti
c
l
as
s
i
f
i
c
ati
o
n.
T
ab
l
e 1
.
C
om
pa
r
i
s
on
of
P
r
op
os
e
d C
l
us
teri
ng
M
e
t
h
o
d
P
r
e
c
i
s
ion
R
e
c
a
ll
F
-
m
e
a
s
u
r
e
P
r
o
p
o
s
e
d
I
FC
M
9
5
.
7
9
3
.
2
9
4
.
4
3
FC
M
7
7
.
9
75
7
6
.
4
2
K
-
m
e
a
n
s
6
9
.
0
9
6
7
.
8
1
6
8
.
4
4
T
ab
l
e 2
.
C
om
pa
r
i
s
on
of
P
r
o
po
s
ed
Cl
as
s
i
f
i
c
ati
on
i
n T
erm
s
A
da
pti
v
e C
l
us
ter
i
ng
A
lgo
r
it
h
m
C
las
s
i
f
i
c
a
t
ion
E
r
r
o
r
P
e
r
c
e
n
t
a
g
e
SVM
2
6
.
9
9
A
n
t
i
-
B
a
y
e
s
M
u
lt
i
C
las
s
i
f
i
c
a
t
ion
1
5
.
9
9
P
r
o
p
o
s
e
d
(
M
S
V
M
)
1
3
.
1
3
Evaluation Warning : The document was created with Spire.PDF for Python.