I
nte
rna
t
io
na
l J
o
urna
l o
f
Adv
a
nces in Applie
d Science
s
(
I
J
AAS)
Vo
l.
1
5
,
No
.
3
,
Sep
tem
b
er
20
2
6
,
p
p
.
1
1
3
1
~
1
1
4
6
I
SS
N:
2252
-
8
8
1
4
,
DOI
:
1
0
.
1
1
5
9
1
/ijaas
.
v
1
5
.
i
3
.
pp
1
1
3
1
-
1
1
4
6
1131
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
a
s
.
ia
esco
r
e.
co
m
O
ptimi
za
tion
-
ena
bled ma
chine
lear
ning
wit
h
f
ea
tu
re
selection
for phishin
g
d
etec
tion sy
stem
L
uk
m
a
n Adeba
y
o
O
g
un
dele
1
,
J
uli
us
T
em
it
a
y
o
Adepo
j
u
2
,
F
em
i Em
ma
nu
el
Ay
o
3
,
I
da
y
a
t
Abik
e
Aka
no
3
,
Abid
em
i Em
m
a
nu
el
Adeniy
i
4
,
H
a
lleluy
a
h
O
luwa
t
o
bi Awo
rinde
5
,
O
luwa
s
eg
un
J
uli
us
Aro
ba
6
,7
,
G
bo
hu
nm
i A
j
ibes
i
n
8
1
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
F
a
c
u
l
t
y
o
f
S
c
i
e
n
c
e
,
O
l
a
b
i
s
i
O
n
a
b
a
n
j
o
U
n
i
v
e
r
s
i
t
y
,
A
g
o
-
I
w
o
y
e
,
N
i
g
e
r
i
a
2
D
e
p
a
r
t
me
n
t
o
f
M
a
t
h
e
m
a
t
i
c
a
l
S
c
i
e
n
c
e
,
F
a
c
u
l
t
y
o
f
P
h
y
s
i
c
a
l
S
c
i
e
n
c
e
,
U
n
i
v
e
r
si
t
y
o
f
I
l
o
r
i
n
,
I
l
o
r
i
n
,
N
i
g
e
r
i
a
3
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
F
a
c
u
l
t
y
o
f
C
o
m
p
u
t
i
n
g
,
U
n
i
v
e
r
s
i
t
y
o
f
I
l
e
s
a
,
I
l
e
sa
,
N
i
g
e
r
i
a
4
D
e
p
a
r
t
me
n
t
o
f
I
n
f
o
r
mat
i
o
n
S
y
s
t
e
ms
,
F
a
c
u
l
t
y
o
f
C
o
m
p
u
t
i
n
g
S
c
i
e
n
c
e
a
n
d
En
g
i
n
e
e
r
i
n
g
,
O
b
a
f
e
m
i
A
w
o
l
o
w
o
U
n
i
v
e
r
si
t
y
,
I
l
e
-
I
f
e
,
N
i
g
e
r
i
a
5
D
e
p
a
r
t
m
e
n
t
o
f
I
n
f
o
r
m
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
,
F
a
c
u
l
t
y
o
f
A
c
c
o
u
n
t
i
n
g
a
n
d
I
n
f
o
r
m
a
t
i
c
s
,
D
u
r
b
a
n
U
n
i
v
e
r
s
i
t
y
o
f
T
e
c
h
n
o
l
o
g
y
,
D
u
r
b
a
n
,
S
o
u
t
h
A
f
r
i
c
a
6
D
e
p
a
r
t
me
n
t
o
f
O
p
e
r
a
t
i
o
n
s
a
n
d
Q
u
a
l
i
t
y
M
a
n
a
g
e
m
e
n
t
,
F
a
c
u
l
t
y
o
f
M
a
n
a
g
e
m
e
n
t
S
c
i
e
n
c
e
s
,
D
u
r
b
a
n
U
n
i
v
e
r
s
i
t
y
o
f
T
e
c
h
n
o
l
o
g
y
,
D
u
r
b
a
n
,
S
o
u
t
h
A
f
r
i
c
a
7
C
e
n
t
r
e
f
o
r
E
c
o
l
o
g
i
c
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
F
a
c
u
l
t
y
o
f
E
n
g
i
n
e
e
r
i
n
g
a
n
d
B
u
i
l
t
E
n
v
i
r
o
n
m
e
n
t
,
U
n
i
v
e
r
s
i
t
y
o
f
J
o
h
a
n
n
e
s
b
u
r
g
,
J
o
h
a
n
n
e
s
b
u
r
g
,
S
o
u
t
h
A
f
r
i
c
a
8
N
H
S
Ed
u
c
a
t
i
o
n
f
o
r
S
c
o
t
l
a
n
d
,
E
d
i
n
b
u
r
g
h
,
U
n
i
t
e
d
K
i
n
g
d
o
m
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
n
1
4
,
2
0
2
5
R
ev
is
ed
J
u
l 2
9
,
2
0
2
6
Acc
ep
ted
Au
g
1
3
,
2
0
2
6
T
h
e
ra
p
i
d
e
v
o
l
u
ti
o
n
o
f
p
h
is
h
i
n
g
m
e
th
o
d
s
h
a
s
l
o
n
g
b
e
e
n
a
th
re
a
t
t
o
c
y
b
e
rse
c
u
r
i
ty
t
h
a
t
d
e
m
a
n
d
s a
d
a
p
t
i
v
e
a
n
d
i
n
te
ll
i
g
e
n
t
d
e
t
e
c
t
i
o
n
tec
h
n
i
q
u
e
s
.
T
h
i
s
c
u
r
re
n
t
wo
r
k
e
n
v
is
i
o
n
s
a
n
im
p
ro
v
e
d
p
h
is
h
i
n
g
d
e
tec
ti
o
n
s
y
s
tem
th
a
t
i
n
v
o
l
v
e
s
th
e
i
n
te
g
ra
t
i
o
n
o
f
m
a
c
h
i
n
e
lea
r
n
i
n
g
(
M
L
)
w
i
th
f
e
a
t
u
re
s
e
lec
t
io
n
f
o
r
i
m
p
r
o
v
i
n
g
th
e
p
e
rf
o
rm
a
n
c
e
o
f
re
a
l
-
t
ime
c
l
a
ss
i
fica
t
io
n
.
A
c
o
m
p
re
h
e
n
s
i
v
e
d
a
tas
e
t
wa
s
o
p
t
im
ize
d
a
n
d
p
re
p
r
o
c
e
ss
e
d
t
h
r
o
u
g
h
Ch
i
-
s
q
u
a
re
(C
HI
)
f
e
a
t
u
re
s
e
lec
t
i
o
n
t
o
re
d
u
c
e
d
i
m
e
n
si
o
n
a
li
t
y
w
h
i
le
p
re
se
r
v
i
n
g
t
h
e
m
o
st
d
isc
rim
i
n
a
ti
v
e
fe
a
tu
re
s
.
Ra
n
d
o
m
f
o
re
s
t
(
R
F
)
a
l
g
o
r
i
t
h
m
,
se
lec
te
d
f
o
r
i
ts
r
o
b
u
st
n
e
ss
,
i
n
te
r
p
re
ta
b
il
it
y
,
a
n
d
h
i
g
h
c
o
m
p
u
ta
ti
o
n
a
l
e
ff
ic
ie
n
c
y
f
o
r
b
i
n
a
r
y
c
la
ss
i
fic
a
t
i
o
n
,
wa
s
e
m
p
l
o
y
e
d
a
n
d
a
c
h
ie
v
e
d
a
n
a
c
c
u
r
a
c
y
o
f
9
6
.
2
5
%
,
e
ff
e
c
t
i
v
e
l
y
i
d
e
n
ti
f
y
i
n
g
s
o
p
h
i
st
ica
te
d
p
h
i
s
h
i
n
g
a
tt
a
c
k
s.
T
h
e
a
rc
h
i
tec
t
u
r
e
is
d
e
s
i
g
n
e
d
t
o
o
p
e
ra
t
e
o
n
la
rg
e
d
a
tas
e
ts
wit
h
m
i
n
ima
l
c
o
m
p
u
ta
ti
o
n
a
l
o
v
e
r
h
e
a
d
b
u
t
wi
t
h
h
i
g
h
p
re
c
i
si
o
n
a
n
d
sc
a
la
b
il
it
y
.
E
x
p
e
r
ime
n
ta
l
re
s
u
l
ts
a
ls
o
i
n
d
ic
a
t
e
t
h
e
d
e
p
e
n
d
a
b
il
it
y
o
f
t
h
e
s
y
s
te
m
w
i
t
h
a
n
a
re
a
u
n
d
e
r
t
h
e
c
u
r
v
e
(
AUC
)
-
re
c
e
i
v
e
r
o
p
e
ra
ti
n
g
c
h
a
r
a
c
te
ri
st
ic
(
ROC
)
sc
o
re
o
f
1
.
0
,
i
n
d
i
c
a
ti
n
g
p
e
r
fe
c
t
se
p
a
ra
ti
o
n
o
f
p
h
i
s
h
i
n
g
f
r
o
m
n
o
rm
a
l
c
a
se
s
.
I
n
a
d
d
i
ti
o
n
,
a
n
a
d
a
p
ti
v
e
c
las
si
fic
a
t
i
o
n
m
o
d
e
l
wa
s
u
se
d
t
o
a
c
c
o
m
m
o
d
a
t
e
u
n
k
n
o
wn
d
a
ta
se
ts
wit
h
a
1
0
0
%
a
c
c
u
ra
c
y
sc
o
re
.
T
h
e
re
s
u
lt
s
a
b
o
v
e
id
e
n
t
if
y
t
h
e
p
o
te
n
t
ial
o
f
a
d
a
p
ti
v
e
,
fe
a
t
u
re
-
o
p
t
imiz
e
d
M
L
sy
s
tem
s
i
n
th
e
a
b
il
it
y
t
o
h
a
n
d
l
e
d
y
n
a
m
ic
p
h
i
s
h
i
n
g
a
t
tac
k
s
a
n
d
e
n
a
b
le
p
r
o
a
c
t
iv
e
c
y
b
e
rse
c
u
ri
t
y
d
e
fe
n
se
.
K
ey
w
o
r
d
s
:
Ad
ap
tiv
e
C
y
b
er
cr
im
e
C
y
b
er
s
ec
u
r
ity
Featu
r
e
s
elec
tio
n
Ma
ch
in
e
lear
n
in
g
Ph
is
h
in
g
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Ab
id
em
i E
m
m
an
u
el
Ad
en
iy
i
Dep
ar
tm
en
t o
f
I
n
f
o
r
m
atio
n
Sy
s
tem
s
,
Facu
lty
o
f
C
o
m
p
u
tin
g
Scien
ce
an
d
E
n
g
in
ee
r
in
g
Ob
af
em
i A
wo
lo
wo
Un
iv
e
r
s
ity
I
le
-
I
f
e,
Nig
er
ia
E
m
ail:
ae
ad
en
iy
i@
o
au
if
e.
ed
u
.
n
g
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
cy
b
e
r
s
ec
u
r
ity
th
r
ea
t
en
v
ir
o
n
m
en
t
h
as
b
ec
o
m
e
m
o
r
e
c
o
m
p
licated
in
r
ec
e
n
t
y
ea
r
s
,
a
n
d
o
n
e
o
f
th
e
m
o
s
t
p
r
ev
alen
t
a
n
d
d
estru
ctiv
e
ty
p
es
o
f
cy
b
e
r
cr
im
e
is
p
h
is
h
in
g
.
T
h
e
ef
f
o
r
t
to
s
teal
p
r
iv
ate
in
f
o
r
m
atio
n
,
in
clu
d
in
g
p
ass
wo
r
d
s
,
u
s
er
n
a
m
es,
an
d
f
in
an
cial
in
f
o
r
m
ati
o
n
,
b
y
p
o
s
in
g
as
a
r
eliab
le
s
o
u
r
ce
in
elec
tr
o
n
ic
co
m
m
u
n
icatio
n
s
is
k
n
o
wn
as
p
h
is
h
in
g
.
T
o
tak
e
ad
v
an
tag
e
o
f
b
o
th
tec
h
n
o
lo
g
ical
an
d
h
u
m
an
s
h
o
r
tc
o
m
in
g
s
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
8
1
4
I
n
t J Ad
v
Ap
p
l Sci
,
Vo
l.
1
5
,
No
.
3
,
Sep
tem
b
er
2
0
2
6
:
1
1
3
1
-
1
1
4
6
1132
attac
k
er
s
em
p
lo
y
a
v
a
r
iety
o
f
d
ec
ep
tio
n
tec
h
n
iq
u
es
to
tr
i
ck
p
eo
p
le
in
t
o
u
n
i
n
ten
tio
n
all
y
d
is
clo
s
in
g
p
r
iv
ate
in
f
o
r
m
atio
n
.
As
a
p
o
p
u
la
r
en
tr
y
p
o
i
n
t
f
o
r
a
v
ar
iety
o
f
c
y
b
er
d
an
g
er
s
,
p
h
is
h
in
g
was
f
o
u
n
d
t
o
b
e
r
esp
o
n
s
ib
le
f
o
r
o
v
er
3
0
% o
f
all
b
r
ea
ch
es,
ac
co
r
d
in
g
to
th
e
Ver
izo
n
Data
B
r
ea
ch
I
n
v
esti
g
atio
n
s
R
ep
o
r
t
(
2
0
2
1
)
[
1
]
–
[
3
]
.
Ph
is
h
in
g
d
etec
tio
n
m
eth
o
d
s
tr
ad
itio
n
ally
u
s
e
r
u
le
-
b
ased
tech
n
iq
u
es
to
id
en
tify
k
n
o
wn
in
d
icato
r
s
o
f
p
h
is
h
in
g
m
ess
ag
es.
Ho
wev
er
,
th
ese
s
y
s
tem
s
ar
e
s
tatic
an
d
in
ef
f
ec
tiv
e
ag
ain
s
t
e
v
o
lv
in
g
p
h
is
h
in
g
tactics.
Ph
is
h
in
g
attac
k
er
s
ca
n
ea
s
i
ly
b
y
p
ass
r
u
le
-
b
ased
d
ef
e
n
s
es
b
y
cr
ea
tin
g
em
ails
m
i
m
ick
in
g
leg
itima
te
co
m
m
u
n
icatio
n
s
.
T
h
is
lead
s
t
o
a
g
r
o
win
g
d
em
an
d
f
o
r
m
o
r
e
ad
v
an
ce
d
,
in
tellig
en
t,
an
d
ad
ap
tiv
e
d
etec
tio
n
s
y
s
tem
s
.
On
e
s
o
lu
tio
n
is
m
ac
h
in
e
lear
n
in
g
(
ML
)
,
wh
ich
ca
n
r
ec
o
g
n
ize
in
tr
icate
p
atter
n
s
th
at
h
u
m
an
an
aly
s
t
wo
u
ld
n
o
t
im
m
ed
iately
r
ec
o
g
n
ize
.
L
ar
g
e
d
atasets
ca
n
b
e
u
s
ed
to
tr
ain
ML
m
o
d
els,
wh
ic
h
ca
n
ev
alu
ate
a
wid
e
r
an
g
e
o
f
p
ar
am
eter
s
to
d
if
f
er
en
tiate
au
th
en
tic
co
m
m
u
n
icat
io
n
s
f
r
o
m
p
h
is
h
in
g
ef
f
o
r
ts
.
R
esear
ch
h
as
s
h
o
w
n
th
at
s
u
p
er
v
is
ed
lear
n
in
g
m
o
d
els,
s
u
ch
as
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
(
SVM)
,
d
ec
is
io
n
tr
ee
(
DT
)
,
an
d
n
eu
r
al
n
etwo
r
k
(
NN)
,
ca
n
d
etec
t
p
h
is
h
in
g
em
ails
with
h
ig
h
ac
cu
r
ac
y
[
4
]
–
[
8
]
.
Ph
is
h
in
g
r
em
ain
s
t
h
e
m
o
s
t
p
r
ev
alen
t
v
ec
t
o
r
f
o
r
cy
b
e
r
attac
k
s
o
n
in
d
iv
id
u
als,
o
r
g
an
izat
io
n
s
,
an
d
cr
itical
in
f
r
astru
ctu
r
e.
I
ts
ef
f
ec
tiv
en
ess
is
d
u
e
to
th
e
ab
ilit
y
o
f
attac
k
er
s
to
r
ap
id
ly
a
d
ap
t th
ei
r
ap
p
r
o
ac
h
,
b
y
p
ass
tr
ad
itio
n
al
f
ilter
s
,
an
d
ex
p
l
o
it h
u
m
an
an
d
s
y
s
tem
v
u
ln
er
a
b
ilit
ies.
W
i
th
m
o
r
e
o
r
g
a
n
izatio
n
s
d
ep
en
d
i
n
g
o
n
cl
o
u
d
co
m
p
u
tin
g
,
r
em
o
te
wo
r
k
,
an
d
d
ec
en
tr
alize
d
d
i
g
ital
in
f
r
astru
ctu
r
e,
tr
ad
itio
n
al
f
ilter
-
b
ased
m
eth
o
d
s
f
o
r
p
h
is
h
in
g
attac
k
d
etec
tio
n
h
a
v
e
n
ev
er
b
ee
n
in
g
r
ea
ter
d
em
an
d
.
T
h
e
p
r
o
p
o
s
ed
o
p
tim
izatio
n
-
aid
ed
ML
p
latf
o
r
m
d
ir
ec
tly
ad
d
r
ess
es
th
is
n
ee
d
b
y
p
r
o
v
id
in
g
a
d
ep
lo
y
a
b
le
f
r
am
ew
o
r
k
th
at
ac
h
iev
es
an
ac
c
u
r
ac
y
-
e
f
f
icien
cy
tr
ad
eo
f
f
.
I
n
en
ter
p
r
is
e
en
v
ir
o
n
m
en
ts
,
th
e
p
latf
o
r
m
ca
n
b
e
in
s
talled
in
clo
u
d
-
b
ased
s
ec
u
r
e
em
ail
g
atew
ay
s
f
o
r
b
u
lk
f
ilter
in
g
o
f
in
co
m
in
g
a
n
d
o
u
tg
o
in
g
m
ess
ag
es
with
n
eg
lig
ib
le
laten
cy
.
At
th
e
en
d
p
o
in
t,
it
ca
n
b
e
em
b
e
d
d
ed
as
a
p
lu
g
in
o
r
b
r
o
wser
ex
ten
s
io
n
an
d
o
f
f
e
r
r
ea
l
-
tim
e
p
r
o
tectio
n
a
g
ain
s
t
m
alicio
u
s
u
n
if
o
r
m
r
eso
u
r
ce
l
o
ca
to
r
s
(
UR
L
s
)
an
d
p
h
is
h
in
g
s
ites
wh
en
th
e
u
s
er
in
ter
ac
ts
.
T
h
e
lig
h
tweig
h
t
f
ea
tu
r
e
s
elec
tio
n
p
h
ase
is
attain
ed
to
g
u
ar
a
n
tee
th
at
m
o
d
els
c
an
also
b
e
u
s
ed
in
en
v
ir
o
n
m
en
ts
wh
er
e
r
eso
u
r
ce
s
ar
e
lim
ited
,
s
u
ch
as
o
n
m
o
b
il
e
d
ev
ices
o
r
in
ter
n
et
o
f
th
i
n
g
s
g
atew
ay
s
,
to
allo
w
d
ec
en
tr
alize
d
e
n
f
o
r
ce
m
en
t
o
f
s
ec
u
r
ity
.
T
h
r
ea
t
m
itig
atio
n
p
er
s
p
ec
tiv
e,
f
le
x
ib
ilit
y
is
m
o
s
t
im
p
o
r
tan
t
i
n
th
e
s
y
s
tem
.
T
h
r
o
u
g
h
f
e
atu
r
e
s
el
ec
tio
n
an
d
o
p
tim
izatio
n
,
th
e
m
o
d
el
r
esis
ts
o
v
er
f
itti
n
g
to
f
ix
ed
d
atasets
an
d
d
y
n
am
ically
r
etr
ain
s
o
n
n
ew
s
am
p
les
o
f
p
h
is
h
in
g
,
alwa
y
s
s
tay
in
g
s
tr
o
n
g
ag
ai
n
s
t
ev
o
lv
in
g
attac
k
v
ec
to
r
s
.
I
ts
in
teg
r
ab
ilit
y
with
th
r
ea
t
i
n
tellig
en
ce
p
latf
o
r
m
s
(
T
I
Ps
)
also
e
n
h
an
ce
s
co
llectiv
e
d
ef
e
n
s
e:
th
e
m
o
d
el
ca
n
p
r
o
ce
s
s
r
ea
l
-
tim
e
in
d
icato
r
s
o
f
co
m
p
r
o
m
is
e
(
I
o
C
s
)
wh
ile
co
n
tr
ib
u
ti
n
g
b
ac
k
n
ewly
lea
r
n
ed
p
h
is
h
i
n
g
s
ig
n
atu
r
es
to
th
e
o
v
er
all
s
ec
u
r
ity
ec
o
s
y
s
tem
.
T
h
e
im
p
lem
en
tatio
n
o
f
ML
i
n
r
ea
l
-
tim
e
p
h
is
h
in
g
d
etec
tio
n
f
ac
es
ch
allen
g
es
s
u
ch
as
h
ig
h
ac
cu
r
ac
y
an
d
lo
w
laten
cy
,
p
ar
ticu
la
r
ly
wh
en
d
ea
lin
g
with
lar
g
e
-
s
ca
l
e
d
ata.
R
esear
ch
er
s
ar
e
ex
p
lo
r
in
g
ad
v
an
ce
d
ML
tech
n
iq
u
es lik
e
d
ee
p
lear
n
in
g
an
d
en
s
em
b
le
lear
n
in
g
to
en
h
a
n
ce
r
esil
ien
ce
ag
ain
s
t e
v
asiv
e
p
h
is
h
in
g
s
tr
ateg
ies.
T
h
e
f
in
an
cial
a
n
d
r
e
p
u
tatio
n
al
im
p
ac
ts
o
f
p
h
is
h
in
g
attac
k
s
a
r
e
s
ig
n
if
ican
t,
as
th
ey
o
f
ten
s
er
v
e
as
en
tr
y
p
o
in
ts
f
o
r
m
o
r
e
s
o
p
h
is
ticated
cy
b
er
t
h
r
ea
ts
.
As
p
h
is
h
in
g
tactics
ev
o
lv
e,
th
er
e
is
a
n
ee
d
f
o
r
d
etec
ti
o
n
s
y
s
tem
s
th
at
ca
n
id
en
tify
k
n
o
wn
attac
k
s
an
d
d
etec
t
n
o
v
el
an
d
em
er
g
i
n
g
f
o
r
m
s
in
r
ea
l
tim
e.
T
h
is
s
tu
d
y
a
im
s
to
d
ev
elo
p
a
n
ML
-
b
a
s
e
d
p
h
i
s
h
i
n
g
d
e
t
e
c
ti
o
n
s
y
s
t
e
m
t
h
a
t
i
n
t
e
g
r
at
e
s
c
y
b
e
r
s
e
c
u
r
i
t
y
v
u
l
n
e
r
a
b
i
li
t
y
f
r
a
m
e
wo
r
k
s
t
o
e
n
h
a
n
c
e
i
t
s
a
d
a
p
t
i
v
e
c
a
p
a
b
i
l
it
i
es
[
9
]
–
[
1
2
]
.
T
h
e
n
o
v
e
l
t
y
o
f
t
h
e
p
r
o
p
o
s
e
d
w
o
r
k
l
i
e
s
i
n
c
o
m
b
i
n
i
n
g
o
p
t
i
m
iz
a
t
i
o
n
-
b
a
s
e
d
f
e
a
t
u
r
e
s
e
l
e
ct
i
o
n
wi
t
h
a
n
a
d
a
p
ti
v
e
ML
m
o
d
e
l
,
w
h
i
c
h
c
o
l
l
e
ct
i
v
e
l
y
e
n
h
a
n
c
e
s
p
r
e
c
is
i
o
n
,
r
e
d
u
c
es
c
o
m
p
u
t
a
t
i
o
n
al
l
o
a
d
,
a
n
d
e
n
a
b
l
e
s
r
e
a
l
-
ti
m
e
p
h
i
s
h
i
n
g
a
t
t
a
c
k
i
d
e
n
t
i
f
i
c
a
t
i
o
n
.
U
n
li
k
e
c
u
r
r
e
n
t
s
o
l
u
t
i
o
n
s
t
h
a
t
s
o
l
v
e
e
i
t
h
e
r
e
n
s
e
m
b
l
e
c
l
a
s
s
i
f
ic
a
t
i
o
n
o
r
f
e
a
t
u
r
e
r
e
d
u
c
tio
n
i
n
d
i
v
i
d
u
a
l
l
y
,
o
u
r
a
p
p
r
o
a
c
h
c
o
m
b
i
n
e
s
t
h
e
tw
o
t
o
g
e
t
h
e
r
u
n
d
e
r
o
n
e
f
r
a
m
e
w
o
r
k
.
T
h
is
s
tu
d
y
d
e
v
elo
p
s
a
n
ad
a
p
tiv
e
p
h
is
h
in
g
d
etec
tio
n
s
y
s
tem
with
a
s
tr
o
n
g
f
o
c
u
s
o
n
C
h
i
-
s
q
u
ar
e
(
C
HI
)
f
ea
tu
r
e
s
elec
tio
n
to
en
h
an
ce
ac
cu
r
ac
y
an
d
ef
f
icien
c
y
.
B
y
ap
p
ly
in
g
t
h
e
C
HI
m
eth
o
d
,
t
h
e
s
y
s
tem
s
elec
t
s
th
e
m
o
s
t
r
elev
an
t
f
ea
tu
r
es,
r
ed
u
cin
g
co
m
p
u
tatio
n
al
o
v
e
r
h
ea
d
w
h
ile
im
p
r
o
v
in
g
class
if
icatio
n
p
er
f
o
r
m
an
ce
.
T
h
is
o
p
tim
ized
f
ea
tu
r
e
s
elec
tio
n
,
c
o
m
b
in
ed
with
ML
,
e
n
s
u
r
es
r
e
al
-
tim
e
p
h
is
h
in
g
d
etec
tio
n
wit
h
h
ig
h
er
p
r
ec
is
io
n
an
d
r
esil
ien
ce
ag
ain
s
t
e
v
asiv
e
tactics.
C
o
m
p
ar
ed
to
t
r
ad
itio
n
al
r
u
le
-
b
ased
o
r
s
tan
d
alo
n
e
ML
ap
p
r
o
ac
h
es,
th
is
m
eth
o
d
en
h
a
n
ce
s
d
etec
tio
n
s
p
ee
d
,
s
ca
lab
ilit
y
,
an
d
ef
f
icien
cy
,
co
n
tr
ib
u
tin
g
to
a
m
o
r
e
r
o
b
u
s
t
an
d
ad
a
p
tiv
e
p
h
is
h
in
g
d
etec
tio
n
f
r
am
ew
o
r
k
.
P
h
i
s
h
i
n
g
r
e
m
ai
n
s
a
m
a
j
o
r
cy
b
e
r
s
e
c
u
r
i
t
y
t
h
r
e
at
,
r
e
q
u
i
r
i
n
g
c
o
n
t
i
n
u
o
u
s
a
d
v
a
n
c
e
m
e
n
ts
in
d
e
t
e
c
t
i
o
n
t
e
c
h
n
i
q
u
e
s
.
M
L
m
o
d
e
l
s
h
a
v
e
b
e
c
o
m
e
e
s
s
e
n
ti
a
l
t
o
o
ls
f
o
r
i
d
e
n
t
i
f
y
i
n
g
p
h
i
s
h
i
n
g
a
t
t
e
m
p
ts
,
b
u
t
t
h
e
i
r
e
f
f
e
c
t
i
v
e
n
es
s
l
a
r
g
e
l
y
d
e
p
e
n
d
s
o
n
t
h
e
q
u
a
li
t
y
o
f
s
e
l
e
ct
e
d
f
e
a
t
u
r
es
.
O
p
ti
m
i
z
at
i
o
n
t
e
c
h
n
i
q
u
e
s
,
s
u
c
h
as
g
e
n
e
tic
a
l
g
o
r
i
t
h
m
s
(
G
A
)
,
r
e
i
n
f
o
r
c
e
m
e
n
t
l
e
a
r
n
i
n
g
,
a
n
d
s
wa
r
m
i
n
t
e
l
l
i
g
e
n
c
e
,
h
a
v
e
b
e
e
n
i
n
te
g
r
a
t
e
d
i
n
t
o
f
e
a
t
u
r
e
s
e
l
ec
t
i
o
n
p
r
o
c
e
s
s
es
t
o
e
n
h
a
n
c
e
m
o
d
e
l
a
c
c
u
r
a
c
y
a
n
d
c
o
m
p
u
t
a
t
io
n
a
l
e
f
f
i
c
i
e
n
c
y
.
T
h
is
l
it
e
r
a
t
u
r
e
r
e
v
i
e
w
e
x
p
l
o
r
e
s
r
ec
e
n
t
s
t
u
d
i
e
s
o
n
f
e
a
t
u
r
e
s
e
l
e
ct
i
o
n
a
n
d
o
p
t
i
m
i
z
a
t
i
o
n
-
d
r
i
v
e
n
p
h
i
s
h
in
g
d
e
t
e
c
t
i
o
n
,
h
i
g
h
l
i
g
h
t
i
n
g
t
h
ei
r
s
t
r
e
n
g
t
h
s
,
d
r
aw
b
a
c
k
s
,
a
n
d
e
x
is
t
i
n
g
r
e
s
e
a
r
c
h
g
a
p
s
.
2.
RE
L
AT
E
D
WO
RK
S
T
an
im
u
et
a
l
.
[
1
3
]
e
x
am
in
ed
v
ar
io
u
s
f
ea
t
u
r
e
elim
i
n
atio
n
te
ch
n
iq
u
es,
e
m
p
h
asizin
g
th
e
im
p
o
r
tan
ce
o
f
r
ed
u
cin
g
ir
r
ele
v
an
t
f
ea
tu
r
es
i
n
p
h
is
h
in
g
d
etec
tio
n
.
T
h
eir
s
tu
d
y
d
em
o
n
s
tr
ated
th
at
f
ea
tu
r
e
s
elec
tio
n
en
h
an
ce
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ad
v
Ap
p
l Sci
I
SS
N:
2252
-
8
8
1
4
Op
timiz
a
tio
n
-
en
a
b
led
ma
c
h
in
e
lea
r
n
in
g
w
ith
fea
tu
r
e
s
elec
tio
n
fo
r
p
h
is
h
in
g
…
(
Lu
kma
n
A
d
eb
a
yo
Og
u
n
d
ele
)
1133
ac
cu
r
ac
y
,
b
u
t
it
also
in
tr
o
d
u
ce
d
co
n
ce
r
n
s
r
eg
ar
d
in
g
th
e
p
o
te
n
tial
r
em
o
v
al
o
f
c
r
itical
f
ea
tu
r
es,
lead
in
g
to
a
lo
s
s
in
m
o
d
el
g
e
n
er
aliza
tio
n
a
b
ilit
y
.
Fu
r
th
e
r
m
o
r
e
,
th
e
r
esear
ch
d
o
es
n
o
t
an
aly
ze
h
o
w
f
ea
tu
r
e
s
elec
tio
n
im
p
ac
ts
m
o
d
er
n
ar
ch
itectu
r
es,
s
u
ch
as tr
an
s
f
o
r
m
er
-
b
ased
p
h
is
h
in
g
d
e
tectio
n
m
o
d
els.
Ko
cy
ig
it
et
a
l
.
[
1
4
]
p
r
o
p
o
s
ed
a
GA
to
o
p
tim
ize
th
e
s
elec
tio
n
o
f
r
elev
an
t
p
h
is
h
in
g
f
ea
tu
r
es,
im
p
r
o
v
in
g
ac
cu
r
ac
y
wh
ile
r
e
d
u
cin
g
c
o
m
p
u
tatio
n
al
co
m
p
l
ex
ity
.
Ho
wev
er
,
GA
s
o
f
ten
s
u
f
f
er
s
f
r
o
m
s
lo
w
co
n
v
er
g
en
ce
an
d
r
e
q
u
ir
e
ca
r
e
f
u
l
p
ar
am
ete
r
tu
n
in
g
.
Mo
r
eo
v
er
,
th
e
s
tu
d
y
p
r
i
m
ar
ily
f
o
cu
s
ed
o
n
UR
L
-
b
ased
p
h
is
h
in
g
d
etec
tio
n
,
leav
in
g
a
g
ap
in
ev
alu
atin
g
r
ea
l
-
tim
e
p
h
is
h
in
g
s
ce
n
ar
io
s
.
Saee
d
[
1
5
]
in
tr
o
d
u
ce
d
a
n
o
p
tim
izatio
n
-
d
r
iv
e
n
m
et
h
o
d
th
at
p
r
io
r
itizes
th
e
m
o
s
t
im
p
ac
tf
u
l
f
ea
tu
r
es
f
o
r
p
h
is
h
in
g
d
etec
tio
n
,
o
u
tp
e
r
f
o
r
m
in
g
c
o
n
v
e
n
tio
n
al
b
asel
in
e
m
o
d
els.
Desp
ite
its
ef
f
ec
tiv
en
ess
,
th
e
ap
p
r
o
ac
h
r
elies
o
n
h
ig
h
c
o
m
p
u
tatio
n
al
r
eso
u
r
ce
s
,
m
ak
in
g
r
ea
l
-
wo
r
ld
ad
o
p
tio
n
ch
allen
g
i
n
g
.
Fu
r
th
e
r
m
o
r
e
,
it
d
o
es
n
o
t
ac
co
u
n
t
f
o
r
ad
v
er
s
ar
ial
p
h
is
h
in
g
attac
k
s
,
wh
er
e
attac
k
er
s
m
o
d
if
y
web
s
ite
ch
ar
a
cter
is
tics
to
b
y
p
ass
d
etec
tio
n
s
y
s
tem
s
.
An
o
th
er
n
o
tab
le
f
ea
tu
r
e
s
elec
tio
n
tech
n
iq
u
e
in
v
o
l
v
es
f
u
z
zy
r
o
u
g
h
s
et
th
eo
r
y
[
1
6
]
.
T
h
is
m
eth
o
d
ef
f
ec
tiv
ely
r
e
m
o
v
es
r
ed
u
n
d
an
t
f
ea
tu
r
es,
lead
i
n
g
t
o
im
p
r
o
v
e
d
class
if
icatio
n
ac
cu
r
ac
y
.
Ho
wev
er
,
f
u
zz
y
r
o
u
g
h
s
et
m
o
d
els
o
f
ten
r
e
q
u
ir
e
e
x
ten
s
iv
e
co
m
p
u
tatio
n
,
wh
ic
h
m
a
y
lim
it
s
ca
lab
ilit
y
f
o
r
lar
g
e
d
atasets
.
Ad
d
itio
n
ally
,
th
e
s
tu
d
y
lack
s
a
co
m
p
ar
ativ
e
an
aly
s
is
ag
ain
s
t
d
ee
p
lear
n
in
g
-
b
ased
p
h
is
h
in
g
d
etec
tio
n
m
o
d
els,
cr
ea
tin
g
an
ar
ea
f
o
r
f
u
r
th
e
r
r
esear
ch
.
E
n
s
em
b
le
m
eth
o
d
s
,
s
u
ch
as
th
e
weig
h
ted
en
s
em
b
le
m
o
d
el
,
h
av
e
also
b
ee
n
ex
p
lo
r
ed
.
B
id
ab
ad
i
an
d
W
an
g
[
1
7
]
c
o
m
b
in
ed
m
u
ltip
le
class
if
ier
s
u
s
in
g
a
f
ea
tu
r
e
s
elec
tio
n
m
ec
h
an
is
m
,
ac
h
iev
in
g
s
u
p
er
io
r
ac
cu
r
ac
y
.
Ho
wev
e
r
,
en
s
em
b
le
ap
p
r
o
ac
h
es
g
en
er
all
y
r
eq
u
ir
e
h
ig
h
c
o
m
p
u
tatio
n
al
r
eso
u
r
ce
s
,
m
ak
in
g
r
ea
l
-
tim
e
d
ep
l
o
y
m
en
t
d
if
f
ic
u
lt.
Ad
d
itio
n
ally
,
th
eir
s
tu
d
y
d
o
es
n
o
t
d
if
f
er
e
n
tiate
b
etwe
en
d
if
f
er
e
n
t
ty
p
es
o
f
p
h
is
h
in
g
attac
k
s
,
s
u
ch
as sp
ea
r
p
h
is
h
in
g
a
n
d
UR
L
-
b
ased
p
h
is
h
in
g
.
Sin
g
h
et
a
l
.
[
1
8
]
ap
p
lied
p
a
r
ticle
s
war
m
o
p
tim
izatio
n
(
PSO
)
f
o
r
f
ea
tu
r
e
s
elec
tio
n
in
p
h
is
h
in
g
d
etec
tio
n
,
r
ep
o
r
tin
g
h
ig
h
e
r
ac
cu
r
ac
y
.
Desp
ite
its
ad
v
a
n
tag
es,
PS
O
m
o
d
els
a
r
e
s
u
s
ce
p
tib
le
to
s
lo
w
co
n
v
er
g
en
ce
wh
en
h
an
d
lin
g
h
ig
h
-
d
im
en
s
io
n
al
d
atasets
.
Ad
d
itio
n
ally
,
th
e
s
tu
d
y
d
o
es
n
o
t
co
m
p
ar
e
PS
O
with
o
th
er
ev
o
lu
tio
n
ar
y
alg
o
r
ith
m
s
,
s
u
ch
as
GA
o
r
an
t c
o
lo
n
y
o
p
ti
m
izatio
n
(
AC
O)
.
Z
ar
a
et
a
l.
[
1
9
]
in
v
esti
g
ated
d
ee
p
lea
r
n
in
g
-
b
ased
f
ea
tu
r
e
s
elec
tio
n
,
s
h
o
win
g
im
p
r
o
v
e
m
en
ts
in
p
h
is
h
in
g
d
etec
tio
n
ac
c
u
r
ac
y
.
Ho
wev
er
,
d
ee
p
lear
n
in
g
m
o
d
els
ty
p
ically
r
eq
u
ir
e
ex
ten
s
iv
e
lab
elled
d
atasets
,
wh
ich
m
ay
n
o
t
alwa
y
s
b
e
av
a
ilab
le.
T
h
e
s
tu
d
y
also
d
o
es
n
o
t
ass
es
s
th
e
ef
f
ec
tiv
en
ess
o
f
d
ee
p
lear
n
in
g
-
b
ased
f
ea
tu
r
e
s
elec
tio
n
in
d
etec
tin
g
z
er
o
-
d
a
y
p
h
is
h
in
g
attac
k
s
,
wh
ic
h
r
em
ain
a
c
r
itical
co
n
ce
r
n
in
cy
b
er
s
ec
u
r
ity
.
W
h
ile
f
ea
tu
r
e
s
elec
tio
n
,
o
p
tim
izatio
n
tech
n
iq
u
es,
an
d
d
if
f
er
en
t
a
d
v
an
ce
d
ML
m
o
d
els
h
av
e
s
ig
n
if
ican
tly
en
h
a
n
ce
d
p
h
is
h
in
g
d
etec
tio
n
s
y
s
tem
s
,
s
ev
er
al
g
ap
s
r
em
ain
in
cu
r
r
en
t
r
es
ea
r
ch
,
as
p
h
is
h
in
g
attac
k
s
co
n
tin
u
e
to
b
e
a
m
ajo
r
cy
b
e
r
s
ec
u
r
ity
th
r
ea
t,
ex
p
lo
itin
g
u
s
er
s
’
tr
u
s
t
to
s
teal
s
en
s
itiv
e
in
f
o
r
m
atio
n
.
Var
io
u
s
tech
n
iq
u
es
h
a
v
e
b
e
en
d
e
v
elo
p
ed
to
d
etec
t
p
h
i
s
h
in
g
attem
p
ts
,
r
an
g
in
g
f
r
o
m
h
eu
r
is
tic
-
b
ased
ap
p
r
o
ac
h
es
to
ML
an
d
d
ee
p
lear
n
in
g
m
o
d
els.
T
h
is
liter
atu
r
e
r
ev
iew
d
is
cu
s
s
es
k
ey
s
tu
d
ies
th
at
p
r
o
p
o
s
e
d
if
f
er
en
t
m
eth
o
d
s
f
o
r
p
h
is
h
in
g
web
s
ite
d
etec
tio
n
.
I
n
th
is
s
tu
d
y
p
r
o
p
o
s
e
a
p
h
is
h
i
n
g
d
etec
tio
n
m
o
d
el
th
at
lev
er
ag
es
th
e
C
HI
f
ea
tu
r
e
s
elec
tio
n
m
eth
o
d
co
m
b
in
ed
with
a
r
an
d
o
m
f
o
r
est
(
C
HI
+
R
F)
clas
s
if
ier
to
en
h
an
ce
d
etec
tio
n
ac
cu
r
ac
y
.
T
h
e
C
HI
m
eth
o
d
e
f
f
ec
tiv
el
y
s
elec
ts
th
e
m
o
s
t
r
elev
an
t
f
ea
tu
r
es,
r
ed
u
cin
g
d
im
en
s
io
n
ality
wh
ile
p
r
eser
v
i
n
g
cr
itical
in
f
o
r
m
atio
n
,
a
n
d
t
h
e
R
F
alg
o
r
ith
m
en
s
u
r
es
r
o
b
u
s
t
class
if
icatio
n
b
y
ag
g
r
eg
atin
g
m
u
ltip
le
DT
s
,
th
e
r
eb
y
im
p
r
o
v
i
n
g
o
v
er
all
p
er
f
o
r
m
an
ce
an
d
r
esil
ien
ce
ag
ain
s
t p
h
is
h
in
g
attac
k
s
.
Ma
n
er
ik
er
et
a
l.
[
2
0
]
t
r
ac
e
th
e
ev
o
lu
tio
n
o
f
p
h
is
h
in
g
UR
L
d
etec
tio
n
f
r
o
m
t
r
ad
itio
n
al
b
lack
lis
t
an
d
r
u
le
-
b
ased
ap
p
r
o
ac
h
es,
wh
ich
wer
e
f
ast
b
u
t
wea
k
in
d
etec
tin
g
n
ew
o
r
s
lig
h
tly
m
o
d
if
ied
a
ttack
s
,
to
class
ica
l
ML
m
o
d
els
t
h
at
r
elied
o
n
m
an
u
ally
c
r
af
ted
le
x
ical
an
d
d
o
m
ain
f
e
atu
r
es.
W
h
ile
R
F
an
d
SVM
alg
o
r
ith
m
s
p
er
f
o
r
m
ed
b
etter
,
th
eir
f
ea
tu
r
e
en
g
in
ee
r
in
g
d
e
p
en
d
e
n
cy
at
th
e
ex
p
e
n
s
e
o
f
f
lex
i
b
ilit
y
to
o
b
f
u
s
ca
te
UR
L
s
was
a
f
law.
Su
b
s
eq
u
en
t
d
ee
p
lear
n
i
n
g
m
eth
o
d
s
u
s
in
g
co
n
v
o
l
u
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NNs
)
an
d
r
ec
u
r
r
en
t
n
eu
r
al
n
etwo
r
k
s
(
R
NNs
)
r
ed
u
ce
d
m
a
n
u
ally
en
g
in
ee
r
e
d
f
ea
tu
r
e
g
e
n
er
atio
n
b
y
d
ir
ec
tly
lear
n
i
n
g
f
r
o
m
r
aw
UR
L
s
,
b
u
t
f
ailed
to
m
o
d
el
lo
n
g
-
ter
m
d
e
p
en
d
en
cies
am
o
n
g
UR
L
co
n
s
titu
en
ts
.
T
h
e
au
th
o
r
s
id
en
tify
th
is
s
h
o
r
tf
all
an
d
ar
g
u
e
th
at
tr
an
s
f
o
r
m
e
r
m
o
d
el
s
,
as
th
ey
ar
e
ca
p
ab
le
o
f
m
o
d
ellin
g
lo
ca
l
an
d
g
lo
b
al
s
eq
u
en
tial
p
atter
n
s
with
atten
tio
n
m
ec
h
an
is
m
s
,
o
f
f
er
a
p
r
o
m
is
in
g
s
o
lu
tio
n
to
th
e
lac
k
o
f
h
an
d
-
cr
af
te
d
f
ea
tu
r
es
in
p
h
is
h
in
g
d
etec
tio
n
,
an
d
th
u
s
th
eir
co
n
tr
ib
u
tio
n
(
UR
L
tr
an
s
f
o
r
m
er
(
UR
L
T
r
an
)
)
is
an
im
p
r
o
v
em
en
t
o
v
er
th
e
cu
r
r
en
t state
o
f
wo
r
k
.
J
o
s
h
u
a
et
a
l
.
[
2
1
]
e
m
p
h
asize
t
h
e
s
h
if
t
f
r
o
m
ce
n
tr
alize
d
p
h
is
h
in
g
d
etec
tio
n
m
o
d
els,
wh
ich
a
r
e
p
lag
u
ed
b
y
p
r
iv
ac
y
a
n
d
s
ca
lab
ilit
y
p
r
o
b
lem
s
,
to
d
is
tr
ib
u
ted
a
n
d
ad
ap
tiv
e
ap
p
r
o
ac
h
es.
T
r
ad
iti
o
n
al
ML
an
d
d
ee
p
lear
n
in
g
-
b
ased
p
h
is
h
in
g
d
etec
tio
n
ap
p
r
o
ac
h
es,
alth
o
u
g
h
ef
f
ec
tiv
e,
r
ely
o
n
ce
n
tr
alize
d
d
at
a
co
llectio
n
an
d
ar
e
th
er
ef
o
r
e
v
u
ln
er
ab
le
to
d
ata
leak
ag
e
a
n
d
less
p
r
ac
tical
in
p
r
iv
ac
y
-
s
en
s
itiv
e
en
v
ir
o
n
m
en
ts
.
T
h
e
wo
r
k
a
d
d
r
ess
es
th
e
g
r
o
win
g
ap
p
licab
ilit
y
o
f
f
ed
er
ated
lear
n
in
g
(
FL)
,
en
a
b
lin
g
jo
in
t
t
r
ain
in
g
a
cr
o
s
s
n
o
d
es
with
o
u
t
r
ev
e
alin
g
r
aw
d
ata,
an
d
co
n
tin
u
al
lear
n
in
g
(
C
L
)
,
en
ab
lin
g
i
n
cr
em
en
t
al
ad
ap
tatio
n
to
em
er
g
i
n
g
p
h
i
s
h
in
g
p
atter
n
s
an
d
av
o
id
in
g
ca
tast
r
o
p
h
ic
f
o
r
g
ettin
g
.
Prio
r
wo
r
k
o
n
p
h
is
h
in
g
d
etec
tio
n
f
o
cu
s
ed
lar
g
el
y
o
n
s
tatic
d
atase
ts
an
d
s
u
f
f
er
ed
f
r
o
m
e
v
o
lv
in
g
attac
k
s
u
r
f
ac
es,
em
p
h
asizin
g
th
e
n
e
ed
f
o
r
a
d
ap
tin
g
m
o
d
els
in
r
ea
l
tim
e.
T
h
e
au
th
o
r
s
also
m
en
tio
n
r
ec
e
n
t
ad
v
a
n
ce
s
in
atten
tio
n
-
b
ased
class
if
ier
s
,
wh
ich
p
r
o
m
o
te
r
o
b
u
s
tn
es
s
b
y
atten
d
in
g
to
p
r
o
m
in
e
n
t
UR
L
o
r
c
o
n
ten
t
f
e
atu
r
es,
f
o
r
im
p
r
o
v
ed
d
etec
tio
n
ac
cu
r
ac
y
u
n
d
er
ad
v
er
s
ar
ial
o
r
d
r
if
tin
g
s
ce
n
ar
i
o
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
8
1
4
I
n
t J Ad
v
Ap
p
l Sci
,
Vo
l.
1
5
,
No
.
3
,
Sep
tem
b
er
2
0
2
6
:
1
1
3
1
-
1
1
4
6
1134
B
y
in
teg
r
atin
g
FL,
C
L
,
an
d
atten
tio
n
m
ec
h
a
n
is
m
s
,
th
e
s
tu
d
y
p
o
s
itio
n
s
its
elf
in
a
n
ew
d
ir
ec
t
io
n
o
f
r
esear
ch
th
at
ad
d
r
ess
es
p
r
iv
ac
y
,
ad
a
p
tab
ilit
y
,
an
d
r
o
b
u
s
tn
ess
s
im
u
ltan
e
o
u
s
ly
—
k
ey
d
em
an
d
s
f
o
r
s
u
c
ce
s
s
f
u
l
r
ea
l
-
wo
r
ld
,
lar
g
e
-
s
ca
le
p
h
is
h
in
g
d
etec
tio
n
s
y
s
tem
s
[
2
2
]
,
[
2
3
]
.
3.
M
E
T
H
O
D
T
h
is
s
ec
tio
n
p
r
esen
ts
a
m
et
h
o
d
o
lo
g
ical
f
r
a
m
ewo
r
k
f
o
r
d
e
v
elo
p
in
g
an
ad
a
p
tiv
e
p
h
is
h
in
g
d
etec
tio
n
s
y
s
tem
u
s
in
g
r
ea
l
-
tim
e
ML
tech
n
iq
u
es.
I
t
aim
s
to
m
itig
ate
th
e
d
y
n
am
ic
n
atu
r
e
o
f
p
h
is
h
in
g
th
r
ea
ts
b
y
co
m
b
in
in
g
tr
ad
itio
n
al
cy
b
e
r
s
ec
u
r
ity
f
r
am
ewo
r
k
s
with
r
e
al
-
tim
e
ML
alg
o
r
ith
m
s
.
T
h
e
s
y
s
tem
an
aly
ze
s
,
p
r
ed
icts
,
an
d
r
ea
cts to
n
ew
p
h
is
h
in
g
p
atter
n
s
.
3
.
1
.
Da
t
a
c
o
llect
io
n a
nd
des
cr
iptio
n
T
h
e
p
h
is
h
in
g
d
etec
tio
n
m
o
d
e
l
is
tr
ain
ed
an
d
e
v
alu
ated
u
s
in
g
two
d
is
tin
ct
d
atasets
.
T
h
e
p
h
is
h
in
g
d
etec
tio
n
d
at
aset
s
er
v
es
as
th
e
p
r
im
ar
y
tr
ain
in
g
d
ataset,
co
m
p
r
is
in
g
2
4
7
,
9
5
0
en
tr
ies
with
4
2
d
is
tin
ct
f
ea
tu
r
es
r
elate
d
to
UR
L
s
an
d
d
o
m
ain
s
.
T
h
is
d
ataset
i
s
s
tr
u
ctu
r
ed
in
a
p
an
d
as
Data
Fra
m
e
f
o
r
m
at,
en
ab
lin
g
ef
f
icien
t
d
ata
m
an
ip
u
latio
n
an
d
an
aly
s
is
[
2
4
]
.
On
th
e
o
th
er
h
an
d
,
th
e
Ph
iu
Sii
l
d
ataset
is
u
s
ed
t
o
test
th
e
m
o
d
el’
s
ad
ap
tiv
e
f
u
n
ctio
n
,
en
s
u
r
in
g
its
r
ea
l
-
tim
e
ap
p
licab
ilit
y
.
T
h
is
d
ataset
co
n
s
is
t
s
o
f
2
3
5
,
7
9
5
en
tr
ies with
5
5
d
is
tin
ct
f
ea
tu
r
es,
also
f
o
cu
s
in
g
o
n
UR
L
an
d
d
o
m
ain
-
r
elate
d
ch
ar
ac
te
r
is
tics
.
L
ik
e
th
e
tr
ain
in
g
d
atas
et,
it is
s
tr
u
ctu
r
ed
in
a
p
an
d
as Da
taFr
am
e
f
o
r
m
at
f
o
r
s
ea
m
less
an
aly
s
is
[
2
5
]
.
3
.
2
.
Chi
-
S
qu
a
re
t
est
f
o
r
f
ea
t
ure
re
lev
a
nce
T
o
ass
ess
f
ea
tu
r
e
r
elev
a
n
ce
,
th
is
s
tu
d
y
u
s
e
d
C
HI
(χ
2
)
t
est,
a
s
tatis
tical
m
eth
o
d
f
o
r
id
en
tify
in
g
ass
o
ciatio
n
s
b
etwe
en
f
ea
tu
r
es a
n
d
th
e
p
h
is
h
in
g
class
if
icatio
n
lab
el.
I
n
th
is
ad
a
p
tiv
e
f
r
am
ew
o
r
k
,
C
HI
test
h
elp
s
in
s
elec
tin
g
f
ea
tu
r
es
th
at
m
o
s
t
ef
f
ec
tiv
ely
ca
p
tu
r
e
t
h
e
b
eh
av
io
r
o
f
p
h
is
h
in
g
th
r
ea
ts
,
s
u
c
h
as
u
n
u
s
u
al
UR
L
len
g
th
s
o
r
d
o
m
ain
a
n
o
m
alies.
T
h
e
C
HI
test
s
tati
s
tic
f
o
r
ea
ch
f
ea
tu
r
e
is
ca
lcu
lated
as
(
1
)
.
2
=
∑
(
−
)
2
(
1
)
W
h
er
e
d
en
o
tes
o
b
s
er
v
ed
f
r
e
q
u
en
cy
an
d
r
ep
r
esen
ts
th
e
ex
p
ec
ted
f
r
eq
u
en
cy
o
f
o
cc
u
r
r
e
n
c
es
u
n
d
er
th
e
ass
u
m
p
tio
n
o
f
in
d
ep
en
d
en
ce
.
T
h
is
f
o
r
m
u
la
id
en
tifie
s
f
ea
t
u
r
es
m
o
s
t
ass
o
ciate
d
with
p
h
is
h
in
g
d
etec
tio
n
,
en
s
u
r
in
g
th
at
o
n
ly
t
h
e
m
o
s
t sig
n
if
ican
t c
y
b
er
s
ec
u
r
ity
-
r
elev
a
n
t f
ea
tu
r
es a
r
e
r
etain
ed
.
3
.
3
.
P
r
o
po
s
ed
m
o
de
l:
Chi
-
s
qu
a
re
+
ra
nd
o
m
f
o
re
s
t
T
h
is
s
tu
d
y
u
s
e
d
th
e
RF
class
i
f
ier
,
a
n
ML
m
o
d
el,
to
c
r
ea
te
an
ad
a
p
tiv
e
p
h
is
h
in
g
d
etec
tio
n
s
y
s
tem
.
T
h
e
ap
p
r
o
ac
h
h
as
d
is
tin
ct
ad
v
an
tag
es
wh
en
it
co
m
es
to
m
an
ag
in
g
cy
b
er
s
ec
u
r
ity
-
r
elate
d
d
u
ties
,
esp
ec
ially
wh
en
it
co
m
es
to
an
aly
zin
g
in
tr
icate
,
r
ea
l
-
tim
e
d
ata
s
tr
ea
m
s
th
at
d
ef
in
e
p
h
is
h
in
g
atte
m
p
ts
.
A
tr
ee
-
b
ased
en
s
em
b
le
lear
n
in
g
tech
n
iq
u
e
t
h
at
is
p
ar
ticu
lar
ly
g
o
o
d
at
m
a
n
ag
in
g
n
o
n
-
lin
ea
r
p
atter
n
s
in
p
h
is
h
in
g
d
ata
is
th
e
RF
cla
s
s
if
ier
.
T
h
e
m
o
d
el
ca
n
id
en
tify
v
ar
io
u
s
p
atter
n
s
o
f
p
h
is
h
in
g
b
eh
av
io
r
b
ec
au
s
e
ea
ch
DT
in
th
e
f
o
r
est
is
tr
ain
ed
o
n
a
s
u
b
s
et
o
f
f
ea
tu
r
e
s
.
B
ec
au
s
e
o
f
its
f
lex
ib
ilit
y
,
i
t
is
id
ea
l
f
o
r
cy
b
e
r
s
ec
u
r
ity
,
a
n
ar
ea
with
a
wid
e
v
ar
iety
o
f
q
u
ick
ly
c
h
an
g
in
g
a
ttack
v
ec
to
r
s
.
Fo
r
an
in
p
u
t
,
ea
ch
tr
ee
(
)
in
an
RF
with
tr
ee
s
p
r
o
v
id
e
a
p
r
ed
ictio
n
.
T
h
e
f
in
al
p
r
ed
ictio
n
is
d
er
iv
ed
b
y
a
g
g
r
e
g
atin
g
th
ese
in
d
iv
id
u
al
tr
ee
p
r
ed
ictio
n
s
in
(
2
)
.
=
1
∑
(
)
(
2
)
T
h
is
en
s
em
b
le
m
eth
o
d
e
n
h
a
n
ce
s
th
e
r
o
b
u
s
tn
ess
o
f
p
h
is
h
in
g
d
etec
tio
n
b
y
av
er
a
g
in
g
p
r
ed
ictio
n
s
ac
r
o
s
s
m
u
ltip
le
DT
s
,
m
ak
in
g
th
e
m
o
d
el
less
s
u
s
ce
p
tib
le
t
o
f
alse
p
o
s
itiv
es
an
d
m
o
r
e
r
esil
ien
t
to
em
er
g
in
g
p
h
is
h
in
g
tactics.
RF
is
a
p
o
wer
f
u
l
an
d
v
er
s
atile
en
s
em
b
le
l
ea
r
n
in
g
m
et
h
o
d
u
s
ed
f
o
r
b
o
th
class
if
icatio
n
an
d
r
eg
r
ess
io
n
task
s
.
I
t
b
u
ild
s
u
p
o
n
th
e
co
n
ce
p
t
o
f
DT
b
y
cr
ea
tin
g
a
“f
o
r
est”
o
f
tr
ee
s
an
d
ag
g
r
eg
atin
g
th
eir
p
r
ed
ictio
n
s
to
im
p
r
o
v
e
ac
c
u
r
ac
y
an
d
c
o
n
tr
o
l
o
v
er
f
itti
n
g
.
T
h
e
k
ey
m
at
h
em
atica
l
co
m
p
o
n
en
ts
an
d
eq
u
atio
n
s
th
at
u
n
d
er
p
in
R
F will d
elv
e
as f
o
llo
ws
.
3
.
4
.
Ra
nd
o
m
f
o
re
s
t
c
o
ns
t
ruct
io
n
R
F
i
s
u
tili
ze
d
as
th
e
co
r
e
class
if
ier
f
o
r
d
etec
tin
g
p
h
is
h
in
g
web
s
ites
b
y
an
aly
zin
g
UR
L
an
d
d
o
m
ain
-
r
elate
d
f
ea
tu
r
es.
T
h
e
m
o
d
el
is
tr
ain
ed
o
n
a
l
ab
elled
d
ataset,
d
is
tin
g
u
is
h
in
g
p
h
is
h
in
g
f
r
o
m
non
-
p
h
is
h
in
g
en
tr
ies
b
ased
o
n
ex
tr
ac
ted
c
h
ar
ac
ter
is
tics
.
T
h
e
class
if
icatio
n
p
r
o
ce
s
s
b
eg
in
s
with
in
itializin
g
an
d
tr
ain
in
g
th
e
RF
m
o
d
el
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ad
v
Ap
p
l Sci
I
SS
N:
2252
-
8
8
1
4
Op
timiz
a
tio
n
-
en
a
b
led
ma
c
h
in
e
lea
r
n
in
g
w
ith
fea
tu
r
e
s
elec
tio
n
fo
r
p
h
is
h
in
g
…
(
Lu
kma
n
A
d
eb
a
yo
Og
u
n
d
ele
)
1135
rf_model
=
RandomForestClassifier(random_state=42) rf_model.fit(X_train, y_train)
rf_pred
=
r
f_
m
o
d
e
l.
p
r
e
di
c
t(
X
_
t
e
s
t
)
T
h
e
m
o
d
el
co
n
s
tr
u
cts
m
u
ltip
l
e
DT
s
(
)
tr
ain
ed
o
n
b
o
o
ts
tr
ap
p
ed
s
u
b
s
ets
o
f
th
e
d
ataset.
Giv
en
a
tr
ain
in
g
d
ataset
.
W
h
er
e
r
e
p
r
e
s
e
n
ts
th
e
f
ea
tu
r
e
s
et
o
f
a
UR
L
,
an
d
is
th
e
co
r
r
esp
o
n
d
in
g
lab
el
(
p
h
is
h
in
g
o
r
non
-
p
h
is
h
in
g
)
,
ea
ch
tr
ee
is
tr
ain
ed
o
n
a
s
am
p
led
s
u
b
s
et
(
)
as d
ef
in
ed
in
(
3
)
.
=
{
(
,
)
}
=
1
(
3
)
A
t
e
a
c
h
n
o
d
e
,
i
n
s
t
e
a
d
o
f
c
o
n
s
i
d
e
r
i
n
g
a
ll
f
e
a
t
u
r
es
p
,
a
r
a
n
d
o
m
s
u
b
s
e
t
o
f
m
f
e
a
t
u
r
es
(
m
<
p
)
i
s
s
e
le
c
t
e
d
f
o
r
s
p
l
i
tt
i
n
g
.
T
h
i
s
r
a
n
d
o
m
n
e
s
s
e
n
h
a
n
c
e
s
m
o
d
e
l
g
e
n
e
r
a
l
i
z
a
ti
o
n
a
n
d
r
e
d
u
c
e
s
o
v
e
r
f
i
t
t
i
n
g
.
T
h
e
p
r
e
d
i
c
t
i
o
n
f
o
r
c
l
a
s
s
i
f
y
i
n
g
p
h
i
s
h
i
n
g
a
n
d
n
o
n
-
p
h
i
s
h
i
n
g
p
r
o
c
e
s
s
e
s
a
g
g
r
e
g
a
te
s
r
es
u
l
ts
f
r
o
m
a
l
l
t
r
e
es
i
n
t
h
e
f
o
r
es
t
,
as
s
h
o
w
n
i
n
(
4
)
.
̂
=
∈
(
0
,
1
)
∑
(
(
)
=
)
=
1
(
4
)
W
h
er
e
is
an
in
d
icato
r
f
u
n
ctio
n
th
at
r
etu
r
n
s
1
if
th
e
ℎ
tr
ee
class
if
ie
s
as
clas
s
(
p
h
is
h
in
g
o
r
non
-
p
h
is
h
in
g
)
,
o
th
e
r
wis
e
0
;
is
th
e
to
tal
n
u
m
b
e
r
of
t
r
e
e
s
;
a
n
d
̂
is
th
e
f
in
al
class
if
icatio
n
l
a
b
e
l
.
B
y
ag
g
r
eg
atin
g
p
r
ed
ictio
n
s
f
r
o
m
m
u
ltip
le
DT
s
,
th
e
R
F
m
o
d
el
e
n
h
an
ce
s
d
etec
tio
n
ac
cu
r
ac
y
,
r
e
d
u
ce
s
v
ar
ian
ce
,
an
d
im
p
r
o
v
es r
o
b
u
s
tn
ess
ag
ain
s
t
a
d
v
er
s
ar
ial
p
h
is
h
in
g
tec
h
n
iq
u
es
.
T
h
is
en
s
em
b
le
ap
p
r
o
ac
h
en
s
u
r
es
th
at
th
e
s
y
s
tem
ef
f
ec
tiv
ely
class
if
ies m
alicio
u
s
UR
L
s
wh
ile
m
ain
tain
in
g
lo
w
f
alse p
o
s
itiv
e
r
ates
,
as
d
is
p
lay
ed
in
Fig
u
r
e
1.
Fig
u
r
e
1
.
R
F
m
o
d
el
a
r
ch
itectu
r
e
A
two
-
s
tag
e
p
h
is
h
in
g
d
etec
tio
n
p
ip
elin
e,
co
n
s
is
tin
g
o
f
u
n
iv
ar
iate
f
ea
tu
r
e
s
elec
tio
n
an
d
en
s
em
b
le
class
if
icatio
n
,
i
s
im
p
lem
en
ted
in
Alg
o
r
ith
m
1
.
Pre
lim
in
ar
y
d
ata
ex
p
lo
r
atio
n
an
d
clea
n
i
n
g
o
p
er
atio
n
s
ar
e
d
o
n
e
f
o
r
i
n
p
u
t
d
ataf
r
a
m
e
co
n
tain
i
n
g
r
o
ws
an
d
f
ea
tu
r
es
,
alo
n
g
with
b
in
ar
y
tar
g
et
lab
el
(
p
h
is
h
in
g
o
r
n
o
t)
.
All m
is
s
in
g
en
tr
ies ar
e
eith
er
f
illed
in
o
r
ze
r
o
-
f
illed
to
en
s
u
r
e
th
e
v
alid
ity
o
f
th
e
n
e
x
t step
s
.
F
e
a
t
u
r
e
s
e
l
e
c
t
i
o
n
i
s
b
a
s
ed
o
n
th
e
C
H
I
t
e
s
t
.
A
f
e
a
t
u
r
e
s
e
l
e
c
t
o
r
i
s
tr
a
i
n
ed
o
n
a
n
d
,
w
i
t
h
t
h
e
r
e
t
en
t
i
o
n
o
f
f
e
a
tu
r
e
s
d
e
m
o
n
s
t
r
a
t
i
n
g
t
h
e
g
r
e
a
t
e
s
t
d
e
p
e
n
d
en
c
e
o
n
th
e
ta
r
g
e
t
la
b
e
l
.
T
h
e
o
b
t
a
i
n
e
d
r
e
d
u
c
e
d
d
a
t
a
f
r
a
m
e
w
i
l
l
th
u
s
h
av
e
t
h
e
s
i
z
e
o
f
.
D
i
s
t
r
ib
u
t
io
n
s
a
n
d
co
r
r
e
l
a
t
i
o
n
s
am
o
n
g
s
e
l
e
c
t
e
d
f
e
a
t
u
r
e
s
a
r
e
p
r
e
s
e
n
t
ed
in
h
i
s
to
g
r
a
m
s
an
d
a
h
e
a
tm
a
p
c
o
r
r
es
p
o
n
d
i
n
g
l
y
,
t
o
v
er
i
f
y
t
h
e
ir
in
f
o
r
m
a
t
i
v
e
n
e
s
s
an
d
l
a
c
k
o
f
m
u
l
t
i
co
l
l
in
e
a
r
i
ty
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
8
1
4
I
n
t J Ad
v
Ap
p
l Sci
,
Vo
l.
1
5
,
No
.
3
,
Sep
tem
b
er
2
0
2
6
:
1
1
3
1
-
1
1
4
6
1136
Data
is
s
p
lit
in
to
tr
ain
in
g
an
d
test
in
g
s
am
p
les
wi
th
p
r
o
p
o
r
ti
o
n
s
o
f
7
0
:3
0
(
,
,
,
)
with
a
f
ix
e
d
s
ee
d
v
alu
e
f
o
r
r
ep
r
o
d
u
ci
b
ilit
y
p
u
r
p
o
s
es.
N
ex
t,
a
n
RF
class
if
ier
is
tr
ain
ed
o
n
with
s
p
ec
if
ied
n
u
m
b
er
o
f
t
r
ee
s
an
d
m
ax
im
al
d
ep
t
h
o
f
ea
ch
tr
ee
.
Af
ter
tr
ain
in
g
,
th
e
RF
o
u
tp
u
ts
th
e
p
r
ed
ictio
n
s
o
n
th
e
h
o
l
d
-
o
u
t
test
in
g
s
am
p
le.
Acc
u
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all
,
a
n
d
F1
-
s
co
r
e
m
etr
ics
a
r
e
ca
lc
u
lated
to
ev
alu
ate
its
p
er
f
o
r
m
an
ce
.
Dis
tr
ib
u
tio
n
p
ie
c
h
ar
ts
an
d
f
r
eq
u
en
cy
d
iag
r
am
s
ar
e
m
ad
e
f
o
r
th
e
o
r
ig
i
n
al
an
d
p
a
r
titi
o
n
e
d
d
atasets
.
T
h
e
alg
o
r
ith
m
r
etu
r
n
s
th
e
tr
ai
n
ed
RF
,
a
lis
t
o
f
n
am
es
o
f
s
e
lecte
d
f
ea
tu
r
es
,
an
d
th
e
f
u
ll
s
et
o
f
m
etr
ics.
W
ith
Alg
o
r
ith
m
1
,
a
r
e
d
u
ce
d
f
ea
tu
r
e
s
p
ac
e
o
f
s
ize
an
d
a
r
eliab
le
p
r
ed
icto
r
in
f
o
r
m
o
f
a
n
e
n
s
em
b
le
m
o
d
el
ca
n
b
e
o
b
tain
ed
in
a
co
m
p
ac
t f
o
r
m
.
Alg
o
r
ith
m
1
.
C
HI
+
RF
:
p
h
is
h
in
g
d
etec
tio
n
m
o
d
el
R
eq
u
ir
e:
a
d
atas
et
df
with
f
ea
t
u
r
es a
n
d
tar
g
et
lab
els,
to
p
k
f
e
atu
r
es to
s
elec
t
E
n
s
u
r
e:
a
tr
ain
ed
RF
m
o
d
el
with
s
elec
ted
f
ea
tu
r
es a
n
d
e
v
alu
a
tio
n
m
etr
ics
1:
Data
in
s
p
ec
tio
n
an
d
p
r
ep
r
o
ce
s
s
in
g
:
2:
L
o
ad
th
e
d
ataset
df
3:
C
h
ec
k
d
ataset
s
h
ap
e,
co
lu
m
n
s
,
an
d
s
tatis
tics
(
e.
g
.
,
m
is
s
in
g
v
a
lu
es,
an
d
u
n
iq
u
e
v
alu
es)
4
:
Han
d
le
m
is
s
in
g
v
alu
es (
e.
g
.
,
f
i
ll with
ze
r
o
o
r
im
p
u
te
v
al
u
es)
5
:
Sp
lit th
e
d
f
in
to
f
ea
tu
r
es
X
an
d
tar
g
et
y
,
wh
er
e
y
r
ep
r
esen
ts
p
h
is
h
in
g
lab
els
6
:
Featu
r
e
s
elec
tio
n
:
C
HI
m
eth
o
d
7
:
I
n
itialize
C
HI
s
elec
to
r
to
ch
o
o
s
e
to
p
k
f
ea
tu
r
es
8
:
Fit
th
e
s
elec
to
r
to
X
an
d
Y
9
:
E
x
tr
ac
t a
n
d
s
to
r
e
th
e
in
d
ices a
n
d
n
am
es o
f
th
e
s
elec
ted
f
ea
tu
r
es
1
0
:
R
ed
u
ce
X
to
in
clu
d
e
o
n
ly
t
h
e
s
elec
ted
f
ea
tu
r
es,
f
o
r
m
i
n
g
Xselected
1
1
:
Vis
u
alize
th
e
r
ed
u
ce
d
f
ea
t
u
r
e
s
et
u
s
in
g
h
is
to
g
r
am
s
an
d
a
co
r
r
elatio
n
h
ea
tm
ap
1
2
:
T
r
ain
-
test
s
p
lit:
1
3
:
Sp
lit
Xselected
an
d
y
in
to
tr
ai
n
in
g
an
d
test
in
g
d
atasets
:
(
Xtr
ain
,
Xtest,
y
tr
ain
,
y
test
)
=
tr
ain
test
s
p
lit(
Xselected
,
y
,
test
s
ize
=
0
.
3
,
r
an
d
o
m
s
tate
=
4
2
)
(
5
)
1
4
:
RF
m
o
d
el
tr
ain
in
g
:
1
5
:
I
n
itialize
th
e
RF
class
if
ier
wit
h
d
esire
d
h
y
p
er
p
ar
am
eter
s
(
e.
g
.
,
n
u
m
b
er
o
f
tr
ee
s
,
m
a
x
d
e
p
th
)
1
6
:
T
r
ain
th
e
m
o
d
el
u
s
in
g
Xtr
ai
n
an
d
y
tr
ain
1
7
:
Mo
d
el
ev
alu
atio
n
:
1
8
:
Pre
d
ict
p
h
is
h
in
g
lab
els
y
p
r
ed
f
o
r
Xtest
1
9
:
C
o
m
p
u
te
p
er
f
o
r
m
a
n
ce
m
etr
ics s
u
ch
as a
cc
u
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
an
d
F1
-
s
co
r
e
2
0
:
Vis
u
aliza
tio
n
an
d
r
esu
lts
:
2
1
:
Vis
u
alize
th
e
class
d
is
tr
ib
u
tio
n
(
e.
g
.
,
leg
itima
te
v
s
p
h
is
h
in
g
)
u
s
in
g
a
p
ie
c
h
ar
t
2
2
:
Plo
t th
e
tr
ain
in
g
an
d
test
in
g
la
b
el
d
is
tr
ib
u
tio
n
s
2
3
:
R
etu
r
n
tr
ain
ed
RF
m
o
d
el,
s
ele
cted
f
ea
tu
r
es,
an
d
ev
alu
atio
n
m
etr
ics
3
.
5
.
E
v
a
lua
t
i
o
n
m
et
rics a
lig
ned wit
h c
y
bersec
urit
y
o
bje
ct
iv
es
Sin
ce
p
h
is
h
in
g
d
etec
tio
n
in
cy
b
er
s
ec
u
r
ity
in
v
o
lv
es
co
n
s
id
er
ab
le
r
is
k
s
,
th
e
s
y
s
tem
’
s
ef
f
icac
y
is
ev
alu
ated
u
s
in
g
a
v
ar
iety
o
f
m
etr
ics,
in
clu
d
in
g
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all
,
F1
-
s
co
r
e,
an
d
s
en
s
it
iv
ity
.
E
ac
h
s
ig
n
al
d
is
p
lay
s
a
d
if
f
er
e
n
t a
s
p
ec
t o
f
th
e
m
o
d
el’
s
r
eliab
ilit
y
i
n
s
p
o
ttin
g
p
h
is
h
in
g
attem
p
ts
.
i)
Acc
u
r
ac
y
:
t
h
is
ca
lcu
lates
th
e
p
er
ce
n
tag
e
o
f
au
th
e
n
tic
an
d
p
h
is
h
in
g
UR
L
s
th
at
ar
e
s
u
cc
ess
f
u
lly
ca
teg
o
r
ized
o
u
t
o
f
all
UR
L
s
.
E
v
en
if
it’s
h
elp
f
u
l,
it
m
ig
h
t
n
o
t
b
e
en
o
u
g
h
o
n
its
o
wn
,
p
ar
ticu
lar
ly
wh
en
th
er
e
ar
e
co
n
s
id
er
ab
ly
m
o
r
e
v
alid
UR
L
s
th
an
p
h
is
h
in
g
o
n
es
,
as g
iv
en
in
(
5
)
.
=
(
+
)
+
+
+
(
5
)
ii)
Pre
cisi
o
n
:
th
is
s
h
o
ws
th
e
p
er
c
en
tag
e
o
f
ac
tu
al
p
h
is
h
in
g
d
ete
ctio
n
s
am
o
n
g
all
p
h
is
h
in
g
-
cla
s
s
if
ied
UR
L
s
.
B
y
m
in
im
izin
g
f
alse p
o
s
itiv
es,
h
ig
h
p
r
ec
is
io
n
lo
wer
s
th
e
p
o
s
s
ib
ilit
y
o
f
m
is
tak
en
ly
r
ep
o
r
tin
g
v
alid
UR
L
s
,
as p
r
esen
ted
in
(
6
)
.
=
+
(
6
)
iii)
R
ec
all:
th
is
in
d
icate
s
wh
at
p
er
ce
n
tag
e
o
f
r
ea
l
p
h
is
h
in
g
UR
L
s
th
e
alg
o
r
ith
m
ac
cu
r
at
ely
d
etec
ts
.
I
n
cy
b
er
s
ec
u
r
ity
s
itu
atio
n
s
,
h
ig
h
r
ec
all
is
ess
en
tial
s
in
ce
it
g
u
ar
an
tees
th
at
p
h
is
h
in
g
UR
L
s
ar
e
id
en
tifie
d
with
litt
le
er
r
o
r
,
as g
iv
en
in
(
7
)
.
=
+
(
7
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ad
v
Ap
p
l Sci
I
SS
N:
2252
-
8
8
1
4
Op
timiz
a
tio
n
-
en
a
b
led
ma
c
h
in
e
lea
r
n
in
g
w
ith
fea
tu
r
e
s
elec
tio
n
fo
r
p
h
is
h
in
g
…
(
Lu
kma
n
A
d
eb
a
yo
Og
u
n
d
ele
)
1137
iv
)
F1
-
s
co
r
e:
wh
en
ass
ess
in
g
m
o
d
els
f
o
r
u
n
b
alan
ce
d
d
atasets
,
th
e
F1
-
s
co
r
e,
wh
ich
co
m
b
in
es
p
r
ec
is
io
n
an
d
r
ec
all,
is
v
er
y
in
s
tr
u
ctiv
e.
A
h
ig
h
F1
-
s
co
r
e
s
h
o
ws
th
at
th
e
m
o
d
el
s
u
cc
ess
f
u
lly
s
tr
ik
es
a
b
alan
ce
b
etwe
en
f
alse p
o
s
itiv
es a
n
d
f
alse n
eg
ati
v
es
,
as sh
o
wn
in
(
8
)
.
1
−
=
2
×
+
+
(
8
)
v)
C
o
n
f
u
s
io
n
m
atr
ix
an
aly
s
is
:
co
n
f
u
s
io
n
m
atr
ices
ar
e
g
en
er
ate
d
to
p
r
o
v
id
e
i
n
s
ig
h
t
in
to
m
o
d
e
l
p
er
f
o
r
m
an
c
e
r
eg
ar
d
in
g
p
h
is
h
in
g
an
d
leg
itima
te
UR
L
cla
s
s
if
ica
tio
n
s
.
E
ac
h
m
atr
ix
o
f
f
er
s
a
d
etailed
b
r
ea
k
d
o
wn
o
f
tr
u
e
p
o
s
itiv
es,
f
alse
p
o
s
itiv
es,
tr
u
e
n
eg
ativ
es,
an
d
f
alse
n
e
g
ativ
e
s
,
allo
win
g
f
o
r
a
n
u
an
ce
d
u
n
d
er
s
tan
d
in
g
o
f
m
o
d
el
er
r
o
r
s
.
An
aly
zin
g
f
alse
p
o
s
itiv
es
is
p
ar
ticu
lar
ly
im
p
o
r
tan
t
in
c
y
b
er
s
ec
u
r
ity
t
o
a
v
o
id
f
lag
g
in
g
leg
itima
te
UR
L
s
in
co
r
r
ec
tly
.
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
h
is
s
ec
tio
n
p
r
esen
ts
th
e
r
es
u
lts
o
f
th
e
RF
m
o
d
el
ap
p
lied
to
th
e
p
h
is
h
in
g
d
etec
tio
n
p
r
o
b
lem
.
T
h
e
p
er
f
o
r
m
an
ce
o
f
th
e
m
o
d
el
is
ev
alu
ated
u
s
in
g
s
ev
er
al
m
etr
ics,
in
clu
d
in
g
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
F1
-
s
co
r
e,
an
d
co
m
p
u
tatio
n
al
ef
f
icien
c
y
.
T
h
e
ex
p
er
im
en
tatio
n
f
o
r
th
e
d
e
v
elo
p
e
d
p
h
is
h
in
g
d
e
tectio
n
s
y
s
tem
was
im
p
lem
en
ted
with
C
HI
f
ea
tu
r
e
s
elec
tio
n
an
d
an
R
F
m
o
d
el.
T
h
e
p
h
is
h
in
g
d
ataset
was
lo
ad
ed
u
s
in
g
th
e
Py
th
o
n
J
u
p
y
ter
No
teb
o
o
k
.
4
.
1
.
Sy
s
t
e
m
s
pecif
ica
t
io
ns
T
h
e
s
y
s
tem
is
d
esig
n
ed
to
r
u
n
ef
f
icien
tly
o
n
a
m
ac
h
i
n
e
eq
u
i
p
p
ed
with
an
I
n
tel
Du
al
-
C
o
r
e
p
r
o
ce
s
s
o
r
clo
ck
ed
at
2
.
2
0
GHz
.
T
o
s
u
p
p
o
r
t
th
e
s
m
o
o
th
e
x
ec
u
tio
n
o
f
p
r
o
ce
s
s
es
an
d
en
h
an
ce
co
m
p
u
ta
tio
n
al
ef
f
icien
cy
,
a
m
in
im
u
m
o
f
1
6
GB
o
f
R
AM
is
r
eq
u
ir
e
d
.
A
d
d
itio
n
ally
,
th
e
s
y
s
tem
s
h
o
u
ld
b
e
in
s
talled
o
n
a
s
to
r
a
g
e
d
e
v
ice
with
at
least 5
0
0
GB
o
f
SS
D,
en
s
u
r
in
g
f
aster
d
ata
r
etr
iev
al
a
n
d
im
p
r
o
v
ed
o
v
e
r
all
p
er
f
o
r
m
a
n
ce
(
Fig
u
r
e
2
)
.
4
.
2
.
St
a
t
is
t
ica
l dis
t
ributio
n o
f
t
he
da
t
a
s
et
Fig
u
r
e
2
s
h
o
ws
th
at
th
er
e
ar
e
1
1
9
,
4
0
9
p
h
is
h
in
g
UR
L
s
in
th
e
d
ataset
,
m
ak
in
g
it
4
8
.
2
%
o
f
th
e
d
at
aset,
with
1
2
8
,
5
4
1
leg
itima
te
UR
L
s
,
m
ak
in
g
u
p
5
1
.
8
%
o
f
th
e
d
ata
s
et.
T
h
e
p
ie
ch
a
r
t
illu
s
tr
ates
th
e
d
iv
is
io
n
b
etwe
e
n
p
h
is
h
in
g
s
ites
an
d
g
o
o
d
s
ites
an
d
d
is
p
lay
s
th
at
g
o
o
d
s
ite
co
n
s
is
t
o
f
1
2
8
,
5
4
1
in
s
tan
ce
s
(
5
1
.
8
%),
wh
ile
p
h
is
h
in
g
s
ites
ac
co
u
n
t
f
o
r
1
1
9
,
4
0
9
in
s
tan
ce
s
(
4
8
.
2
%).
T
h
e
alm
o
s
t
b
a
lan
ce
d
d
iv
is
io
n
is
im
p
o
r
ta
n
t
d
u
r
in
g
tr
ain
i
n
g
ML
s
y
s
tem
s
b
ec
au
s
e
it
av
o
id
s
b
iasi
n
g
to
war
d
s
th
e
lar
g
er
class
an
d
allo
ws
p
r
o
p
e
r
p
e
r
f
o
r
m
an
ce
ev
alu
atio
n
o
n
b
o
th
class
es.
W
h
ile
th
e
p
h
is
h
in
g
attac
k
s
r
em
ain
s
o
m
ewh
at
lo
wer
th
an
th
e
o
r
ig
in
al
s
am
p
le
s
,
th
e
s
m
all
m
ar
g
in
r
ef
lects
th
e
co
n
tin
u
ed
s
ize
o
f
p
h
is
h
in
g
attac
k
s
in
th
e
web
en
v
ir
o
n
m
en
t
an
d
th
e
n
ec
ess
ity
f
o
r
r
o
b
u
s
t
d
etec
tio
n
s
y
s
tem
s
ca
p
ab
le
o
f
ef
f
ec
tiv
ely
d
is
tin
g
u
is
h
in
g
b
etwe
en
th
e
tw
o
class
es.
Fig
u
r
e
2
.
D
is
tr
ib
u
tio
n
o
f
th
e
d
ataset
4
.
3
.
F
e
a
t
ure
s
elec
t
io
n us
ing
Chi
-
s
q
ua
re
T
h
e
f
ea
tu
r
e
s
elec
tio
n
m
eth
o
d
was
u
s
ed
to
ex
tr
ac
t
th
e
m
o
s
t
r
elev
an
t
f
ea
tu
r
es
b
ased
o
n
C
HI
m
etr
ics.
T
h
e
C
HI
f
ea
tu
r
e
s
elec
tio
n
m
et
h
o
d
r
ed
u
ce
d
th
e
n
u
m
b
er
o
f
f
e
atu
r
es
f
r
o
m
4
1
to
3
0
,
d
is
ca
r
d
i
n
g
th
e
r
est
as
s
h
o
wn
in
Fig
u
r
e
3
.
T
h
e
f
ea
tu
r
es
s
el
ec
ted
ar
e
s
h
o
wn
in
T
a
b
le
1
.
T
h
is
m
eth
o
d
is
o
f
th
e
f
ilter
-
b
ased
f
ea
tu
r
e
s
u
b
s
et
s
elec
tio
n
ty
p
e,
wh
er
e
f
ea
tu
r
es
ar
e
r
an
k
ed
in
d
ep
e
n
d
en
t
o
f
th
e
lear
n
in
g
alg
o
r
ith
m
.
I
t
is
co
m
p
u
tatio
n
ally
in
ex
p
en
s
iv
e,
s
ca
lab
le
f
o
r
lar
g
e
d
atasets
,
an
d
h
as
g
o
o
d
p
er
f
o
r
m
a
n
ce
f
o
r
ca
teg
o
r
ical
o
r
d
is
cr
ete
f
ea
t
u
r
es
(
e.
g
.
,
p
h
is
h
in
g
f
ac
to
r
s
lik
e
in
clu
d
in
g
"@
"
in
a
UR
L
,
o
r
u
s
ag
e
o
f
an
I
P
ad
d
r
ess
)
.
Ho
wev
er
,
s
in
ce
χ
2
test
s
f
ea
tu
r
es
in
is
o
latio
n
,
it
d
o
es
n
o
t
tak
e
in
to
ac
co
u
n
t
in
ter
ac
ti
o
n
s
b
etwe
en
f
ea
tu
r
es,
an
d
th
at
is
wh
y
it
is
o
f
ten
co
m
p
lem
en
ted
with
e
m
b
ed
d
e
d
m
eth
o
d
s
(
e.
g
.
,
least
ab
s
o
lu
t
e
s
h
r
in
k
ag
e
an
d
s
elec
tio
n
o
p
e
r
ato
r
o
r
L
ASSO
)
o
r
o
p
tim
izatio
n
m
eth
o
d
s
in
r
o
b
u
s
t p
h
is
h
in
g
d
etec
tio
n
s
y
s
tem
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
8
1
4
I
n
t J Ad
v
Ap
p
l Sci
,
Vo
l.
1
5
,
No
.
3
,
Sep
tem
b
er
2
0
2
6
:
1
1
3
1
-
1
1
4
6
1138
Fig
u
r
e
3
.
C
HI
f
ea
tu
r
e
s
elec
tio
n
to
s
elec
t 3
0
f
ea
t
u
r
e
s
T
ab
le
1
.
Featu
r
es selecte
d
u
s
in
g
C
HI
N
o
.
F
e
a
t
u
r
e
N
o
.
F
e
a
t
u
r
e
1
U
R
L
l
e
n
g
t
h
16
N
u
mb
e
r
of
h
y
p
h
e
n
s
in
t
h
e
d
o
m
a
i
n
2
N
u
mb
e
r
o
f
d
o
t
s i
n
U
R
L
17
H
a
v
i
n
g
sp
e
c
i
a
l
c
h
a
r
a
c
t
e
r
s
in
t
h
e
d
o
m
a
i
n
3
H
a
v
i
n
g
r
e
p
e
a
t
e
d
d
i
g
i
t
s
in
U
R
L
18
N
u
mb
e
r
of
s
p
e
c
i
a
l
c
h
a
r
a
c
t
e
r
s
in
t
h
e
d
o
m
a
i
n
4
N
u
mb
e
r
of
d
i
g
i
t
s
in
U
R
L
19
H
a
v
i
n
g
d
i
g
i
t
s
in
t
h
e
d
o
m
a
i
n
5
N
u
mb
e
r
of
s
p
e
c
i
a
l
c
h
a
r
a
c
t
e
r
s
in
U
R
L
20
N
u
mb
e
r
of
d
i
g
i
t
s
in
t
h
e
d
o
m
a
i
n
6
N
u
mb
e
r
of
h
y
p
h
e
n
s
in
U
R
L
21
H
a
v
i
n
g
r
e
p
e
a
t
e
d
d
i
g
i
t
s
in
t
h
e
d
o
m
a
i
n
7
N
u
mb
e
r
of
s
l
a
s
h
e
s
in
U
R
L
22
N
u
mb
e
r
of
s
u
b
d
o
m
a
i
n
s
8
N
u
mb
e
r
of
q
u
e
st
i
o
n
mark
s
in
U
R
L
23
A
v
e
r
a
g
e
s
u
b
d
o
ma
i
n
l
e
n
g
t
h
9
N
u
mb
e
r
of
e
q
u
a
l
s
in
U
R
L
24
H
a
v
i
n
g
d
i
g
i
t
s
in
a
s
u
b
d
o
m
a
i
n
10
N
u
mb
e
r
of
@
in
U
R
L
25
N
u
mb
e
r
of
d
i
g
i
t
s
in
t
h
e
s
u
b
d
o
m
a
i
n
11
N
u
mb
e
r
of
d
o
l
l
a
r
s
in
U
R
L
26
P
a
t
h
l
e
n
g
t
h
12
N
u
mb
e
r
of
e
x
c
l
a
ma
t
i
o
n
m
a
r
k
s
in
U
R
L
27
H
a
v
i
n
g
q
u
e
r
y
13
N
u
mb
e
r
of
p
e
r
c
e
n
t
in
U
R
L
28
H
a
v
i
n
g
a
n
c
h
o
r
14
D
o
ma
i
n
l
e
n
g
t
h
29
En
t
r
o
p
y
of
U
R
L
15
N
u
mb
e
r
of
d
o
t
s
in
t
h
e
d
o
m
a
i
n
30
En
t
r
o
p
y
of
t
h
e
d
o
m
a
i
n
4
.
4
.
T
ra
in
-
t
est
s
pli
t
T
o
s
ep
ar
ate
th
e
d
ataset
in
to
t
r
ain
in
g
an
d
test
in
g
s
ets,
a
7
0
/3
0
s
p
lit
r
atio
was
u
s
ed
.
I
n
p
ar
ti
cu
lar
,
th
e
m
o
d
el
is
tr
ain
ed
u
s
in
g
7
0
%
o
f
th
e
d
ata,
wh
ich
en
ab
les
it
to
r
ec
o
g
n
ize
a
n
d
ad
j
u
s
t
to
th
e
p
atter
n
s
in
th
e
d
ata.
T
h
e
ac
cu
r
ac
y
an
d
ef
f
icac
y
o
f
th
e
m
o
d
el
m
ay
b
e
ass
ess
ed
o
n
u
n
k
n
o
wn
d
ata
b
y
u
s
in
g
th
e
r
em
ain
in
g
3
0
%
as
a
test
in
g
s
et.
I
n
o
r
d
er
to
g
iv
e
t
h
e
m
o
d
el
a
s
ig
n
if
ican
t
q
u
a
n
tity
o
f
t
r
ain
in
g
d
ata
wh
ile
m
ain
tai
n
in
g
a
s
u
itab
le
p
a
r
t
f
o
r
ac
cu
r
ate
p
er
f
o
r
m
a
n
ce
ev
al
u
atio
n
o
n
f
r
esh
s
am
p
les,
th
is
s
p
lit
r
atio
was
u
s
ed
.
T
h
e
r
esu
lts
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
ar
e
s
h
o
wn
in
T
a
b
le
2
.
T
ab
le
2
.
C
HI
+
R
F
class
if
icati
o
n
r
ep
o
r
t
C
l
a
s
s
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F1
-
s
c
o
r
e
S
u
p
p
o
r
t
L
e
g
i
t
i
m
a
t
e
0
.
9
6
0
.
9
7
0
.
9
6
3
8
5
6
9
P
h
i
s
h
i
n
g
0
.
9
7
0
.
9
5
0
.
9
6
3
5
8
1
6
A
c
c
u
r
a
c
y
0
.
9
6
7
4
3
8
5
M
a
c
r
o
a
v
g
0
.
9
6
0
.
9
6
0
.
9
6
7
4
3
8
5
W
e
i
g
h
t
e
d
a
v
g
0
.
9
6
0
.
9
6
0
.
9
6
7
4
3
8
5
4
.
5
.
M
o
del
perf
o
rma
nce
o
n
ph
is
hin
g
det
ec
t
io
n
T
h
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
C
HI
+
R
F
m
o
d
el
was
ev
alu
ated
u
s
i
n
g
a
test
s
et
co
n
tain
in
g
an
e
q
u
al
n
u
m
b
e
r
o
f
p
h
is
h
in
g
an
d
le
g
itima
te
s
am
p
les.
T
h
e
m
o
d
el
d
em
o
n
s
tr
ated
9
6
.
2
5
%
ac
cu
r
ac
y
,
ef
f
ec
ti
v
ely
d
is
tin
g
u
is
h
in
g
b
etwe
en
p
h
is
h
in
g
a
n
d
le
g
itima
te
em
ail,
m
in
im
izin
g
.
T
h
is
h
i
g
h
ac
c
u
r
ac
y
is
ess
en
tial
in
m
i
n
im
izin
g
th
e
r
is
k
o
f
m
is
class
if
y
in
g
p
h
is
h
in
g
attac
k
s
.
Fo
r
p
r
ec
is
io
n
,
th
e
m
o
d
el
ac
h
iev
ed
9
6
.
2
8
%,
in
d
icatin
g
its
ab
ilit
y
to
co
r
r
ec
tly
id
en
tify
p
h
is
h
in
g
attem
p
ts
wh
ile
k
ee
p
in
g
f
alse
p
o
s
itiv
es
to
a
m
in
im
u
m
.
T
h
e
r
ec
all
was
m
ea
s
u
r
ed
at
9
6
.
2
2
%,
r
ef
lectin
g
th
e
m
o
d
el’
s
ef
f
ec
tiv
en
ess
in
ca
p
tu
r
in
g
ac
t
u
al
p
h
is
h
in
g
in
s
tan
ce
s
.
T
o
p
r
o
v
id
e
a
b
alan
ce
d
ev
alu
atio
n
,
th
e
F1
-
s
co
r
e,
wh
ich
is
th
e
h
ar
m
o
n
ic
m
ea
n
o
f
p
r
ec
is
io
n
an
d
r
ec
all,
was
9
6
.
2
2
%.
T
h
is
m
etr
ic
en
s
u
r
es
th
at
th
e
m
o
d
el
m
ai
n
tain
s
b
o
t
h
h
ig
h
ac
cu
r
ac
y
in
class
if
icatio
n
a
n
d
a
s
tr
o
n
g
a
b
ilit
y
to
d
etec
t
p
h
is
h
i
n
g
attem
p
ts
with
o
u
t
p
r
o
d
u
cin
g
e
x
ce
s
s
iv
e
f
alse p
o
s
itiv
es.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ad
v
Ap
p
l Sci
I
SS
N:
2252
-
8
8
1
4
Op
timiz
a
tio
n
-
en
a
b
led
ma
c
h
in
e
lea
r
n
in
g
w
ith
fea
tu
r
e
s
elec
tio
n
fo
r
p
h
is
h
in
g
…
(
Lu
kma
n
A
d
eb
a
yo
Og
u
n
d
ele
)
1139
A
co
n
f
u
s
io
n
m
atr
i
x
was
g
e
n
e
r
ated
to
v
is
u
alize
th
e
m
o
d
el
’
s
p
er
f
o
r
m
an
ce
in
d
is
tin
g
u
is
h
in
g
p
h
is
h
in
g
f
r
o
m
leg
itima
te
s
am
p
les
,
as
s
h
o
wn
in
Fig
u
r
e
4
.
T
h
is
m
atr
i
x
is
ess
en
tial
f
o
r
u
n
d
er
s
tan
d
in
g
th
e
d
is
tr
ib
u
tio
n
o
f
f
alse
p
o
s
itiv
es
an
d
f
alse
n
eg
at
iv
es,
wh
ich
h
elp
s
g
u
id
e
im
p
r
o
v
em
en
ts
in
th
e
m
o
d
el.
As
s
e
en
in
th
e
co
n
f
u
s
io
n
m
atr
ix
,
th
e
n
u
m
b
er
o
f
f
alse
p
o
s
itiv
es
(
leg
itima
te
em
ails
class
if
ied
as
p
h
is
h
in
g
)
was
lo
w,
in
d
icatin
g
th
e
r
o
b
u
s
tn
ess
o
f
th
e
m
o
d
el.
T
h
e
f
alse
n
eg
ativ
e
r
ate
(
p
h
is
h
i
n
g
em
ails
m
is
class
if
ied
as
leg
itima
te)
was
also
m
in
im
al,
wh
ich
is
v
ital f
o
r
p
r
e
v
en
tin
g
p
h
is
h
in
g
attac
k
s
f
r
o
m
b
y
p
ass
in
g
th
e
d
etec
tio
n
s
y
s
tem
.
Fig
u
r
e
4
.
C
o
n
f
u
s
io
n
m
atr
i
x
f
o
r
C
HI
with
RF
m
o
d
el
4
.
6
.
Co
m
pa
riso
n wit
h o
t
her
m
o
dels
T
o
ev
alu
ate
th
e
ef
f
ec
tiv
en
ess
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el,
a
co
m
p
ar
is
o
n
was
m
ad
e
with
o
th
er
ML
m
o
d
els
co
m
m
o
n
l
y
u
s
ed
in
p
h
is
h
in
g
d
etec
tio
n
,
s
u
c
h
as
R
F,
n
aï
v
e
B
ay
es,
an
d
lo
g
is
tic
r
e
g
r
ess
io
n
.
T
h
e
r
esu
lts
ar
e
s
u
m
m
ar
ized
in
T
a
b
le
3
.
As
s
e
en
in
T
ab
le
3
,
th
e
C
HI
+
R
F
m
o
d
el
o
u
tp
er
f
o
r
m
s
o
th
er
m
o
d
els.
W
h
ile
b
ag
g
e
d
tr
ee
p
er
f
o
r
m
ed
well,
it
ex
h
ib
its
s
lig
h
tly
lo
wer
p
r
ec
is
io
n
a
n
d
r
ec
all
t
h
an
t
h
e
C
HI
+
R
F
m
o
d
el.
Fig
u
r
e
5
d
is
p
lay
s
th
e
co
m
p
ar
is
o
n
with
o
th
er
m
o
d
els.
T
h
e
C
HI
+
R
F
ac
h
iev
es
an
o
u
ts
tan
d
in
g
ar
ea
u
n
d
e
r
th
e
cu
r
v
e
(
AUC)
s
co
r
e
o
f
1
.
0
0
,
d
em
o
n
s
tr
atin
g
p
er
f
ec
t
class
s
ep
ar
atio
n
ca
p
a
b
ilit
ies,
wh
ich
s
u
g
g
ests
it
is
ex
ce
p
tio
n
ally
ef
f
ec
tiv
e
at
r
an
k
in
g
p
r
ed
ictio
n
s
ac
cu
r
ately
.
Desp
ite
th
is
h
ig
h
AUC,
its
ac
cu
r
ac
y
o
f
9
6
.
2
5
% in
d
icate
s
th
at
a
s
m
all
n
u
m
b
e
r
o
f
m
is
class
if
icatio
n
s
o
cc
u
r
wh
e
n
a
s
p
ec
if
ic
d
ec
is
io
n
th
r
esh
o
ld
is
ap
p
lied
,
th
o
u
g
h
it
r
em
ain
s
h
ig
h
ly
r
eliab
le
o
v
er
all.
C
o
m
p
ar
ativ
ely
,
th
e
b
ag
g
ed
DT
m
o
d
el
f
o
llo
ws
clo
s
ely
w
ith
an
AU
C
o
f
0
.
9
9
,
p
r
o
v
id
in
g
r
o
b
u
s
t
p
er
f
o
r
m
an
ce
an
d
g
o
o
d
g
en
er
aliza
b
ilit
y
d
u
e
t
o
r
ed
u
c
e
d
v
ar
ian
ce
.
T
h
e
m
u
lti
-
lay
er
p
er
ce
p
tr
o
n
(
ML
P)
an
d
e
x
tr
em
e
g
r
ad
ien
t
b
o
o
s
tin
g
(
XGBo
o
s
t
)
also
p
er
f
o
r
m
well
with
AUC
s
co
r
es
o
f
0
.
9
6
a
n
d
0
.
9
3
,
r
esp
ec
tiv
ely
,
s
h
o
win
g
th
eir
s
tr
en
g
th
i
n
ca
p
tu
r
in
g
co
m
p
lex
p
atter
n
s
,
al
b
eit
with
p
o
ten
tially
h
i
g
h
er
co
m
p
u
tatio
n
al
d
em
an
d
s
.
L
o
g
is
tic
r
eg
r
ess
io
n
,
with
a
lo
wer
AUC
o
f
0
.
8
7
,
d
em
o
n
s
t
r
ates
s
o
lid
b
u
t
m
o
r
e
lim
ited
p
er
f
o
r
m
an
ce
,
as
it
m
ay
s
tr
u
g
g
le
with
n
o
n
-
lin
ea
r
r
elatio
n
s
h
ip
s
th
at
en
s
em
b
le
m
o
d
els
h
an
d
le
ef
f
ec
tiv
ely
.
I
n
s
u
m
m
ar
y
,
C
HI
+
R
F
s
tan
d
s
o
u
t
as
th
e
b
est
m
o
d
el
b
ased
o
n
AUC,
b
u
t
th
e
ch
o
ice
b
etwe
en
th
ese
m
o
d
els
co
u
ld
d
ep
en
d
o
n
s
p
ec
if
ic
n
ee
d
s
s
u
ch
as
m
o
d
el
in
ter
p
r
etab
ilit
y
,
co
m
p
u
tatio
n
a
l
ef
f
icien
cy
,
an
d
th
e
d
esire
d
b
alan
ce
b
etwe
en
AUC
an
d
a
cc
u
r
ac
y
,
as
s
h
o
wn
in
Fig
u
r
e
6
.
T
ab
le
3
.
Per
f
o
r
m
an
ce
co
m
p
a
r
is
o
n
o
f
class
if
ier
s
C
l
a
s
s
i
f
i
e
r
A
c
c
u
r
a
c
y
(
%
)
P
r
e
c
i
s
i
o
n
(
%
)
R
e
c
a
l
l
(
%
)
F1
-
s
c
o
r
e
(
%
)
S
p
e
c
i
f
i
c
i
t
y
(
%
)
C
H
I
+
RF
9
6
.
2
5
9
6
.
2
8
9
6
.
2
2
9
6
.
2
4
9
7
.
1
0
B
a
g
g
e
d
t
r
e
e
9
6
.
0
1
9
6
.
0
4
9
5
.
9
8
9
6
.
0
1
9
6
.
8
1
Lo
g
i
s
t
i
c
r
e
g
r
e
s
s
i
o
n
8
0
.
2
3
8
0
.
7
7
7
9
.
9
1
8
0
.
0
3
8
7
.
1
3
X
G
B
o
o
s
t
8
6
.
1
4
8
6
.
3
5
8
6
.
0
0
8
6
.
0
7
8
9
.
9
2
M
L
P
8
9
.
3
1
8
9
.
2
9
8
9
.
3
4
8
9
.
3
0
8
8
.
6
4
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
8
1
4
I
n
t J Ad
v
Ap
p
l Sci
,
Vo
l.
1
5
,
No
.
3
,
Sep
tem
b
er
2
0
2
6
:
1
1
3
1
-
1
1
4
6
1140
Fig
u
r
e
5
.
B
ar
ch
a
r
t c
o
m
p
ar
is
o
n
with
o
th
er
m
o
d
els
Fig
u
r
e
6
.
R
ec
eiv
er
o
p
er
atin
g
c
h
ar
ac
ter
is
tic
(
R
OC
)
cu
r
v
e
o
f
a
ll m
o
d
els
4
.
7
.
Ada
ptiv
e
re
t
ra
ini
ng
f
un
ct
io
n
T
h
e
ad
a
p
tiv
e
r
et
r
ain
in
g
f
u
n
ctio
n
is
d
esig
n
ed
t
o
en
h
an
ce
an
ML
m
o
d
el’
s
r
esil
ien
ce
ag
ain
s
t
ev
o
lv
in
g
th
r
ea
ts
,
p
ar
ticu
lar
ly
in
th
e
c
o
n
tex
t
o
f
p
h
is
h
in
g
d
etec
tio
n
.
I
t
ac
co
m
p
lis
h
es
th
is
b
y
allo
win
g
th
e
m
o
d
el
to
in
co
r
p
o
r
ate
n
ew
d
ata
d
y
n
am
i
ca
lly
,
en
s
u
r
in
g
CL
an
d
im
p
r
o
v
in
g
its
p
r
ed
ictiv
e
ca
p
ab
ilit
ies.
Up
o
n
r
ec
eiv
in
g
f
r
esh
d
ata,
th
e
f
u
n
ctio
n
f
ir
s
t
ch
ec
k
s
f
o
r
d
ata
v
alid
ity
an
d
a
p
p
lies
a
p
r
e
-
ex
is
tin
g
f
ea
tu
r
e
s
elec
tio
n
m
eth
o
d
t
o
s
tr
ea
m
lin
e
th
e
in
p
u
t
f
ea
tu
r
es.
I
t
th
en
r
etr
ain
s
th
e
RF
m
o
d
el
u
s
in
g
th
e
u
p
d
ated
d
ataset
an
d
co
m
p
u
tes
v
ar
i
o
u
s
Evaluation Warning : The document was created with Spire.PDF for Python.