I
nd
o
ne
s
ia
n J
o
urna
l o
f
E
lect
rica
l En
g
ineering
a
nd
Co
m
pu
t
er
Science
Vo
l.
4
3
,
No
.
1
,
Ju
ly
2
0
2
6
,
p
p
.
299
-
31
3
I
SS
N:
2
5
0
2
-
4
7
5
2
,
DOI
: 1
0
.
1
1
5
9
1
/ijeecs.v
4
3
.i
1
.
p
p
299
-
31
3
299
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ee
cs.ia
esco
r
e.
co
m
Deep
Q
lea
rn
ing
a
lg
o
rithm f
o
r de
t
ecting D
Do
S attac
ks o
n
Io
T dev
ices
L
a
na
K
a
m
la
Ahm
ed
1
,
K
a
y
h
a
n Z
ra
r
G
ha
f
o
o
r
2
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
a
n
d
I
n
f
o
r
ma
t
i
o
n
T
e
c
h
n
o
l
o
g
y
,
C
o
l
l
e
g
e
o
f
S
c
i
e
n
c
e
,
S
a
l
a
h
a
d
d
i
n
U
n
i
v
e
r
si
t
y
-
Er
b
i
l
,
Er
b
i
l
,
I
r
a
q
2
D
e
p
a
r
t
me
n
t
o
f
I
n
f
o
r
mat
i
o
n
a
n
d
C
o
m
mu
n
i
c
a
t
i
o
n
Te
c
h
n
o
l
o
g
y
En
g
i
n
e
e
r
i
n
g
,
Er
b
i
l
P
o
l
y
t
e
c
h
n
i
c
U
n
i
v
e
r
si
t
y
,
Er
b
i
l
,
I
r
a
q
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Feb
2
1
,
2
0
2
6
R
ev
is
ed
J
u
n
1
0
,
2
0
2
6
Acc
ep
ted
J
u
n
2
7
,
2
0
2
6
Th
e
ra
p
id
e
x
p
a
n
sio
n
o
f
i
n
tern
e
t
o
f
th
i
n
g
s
(Io
T)
n
e
two
r
k
s
h
a
s
h
e
ig
h
ten
e
d
se
c
u
rit
y
risk
s
,
p
a
rt
icu
larly
re
g
a
r
d
in
g
d
istri
b
u
ted
d
e
n
ial
o
f
se
rv
ic
e
(DD
o
S
)
a
tt
a
c
k
s
a
g
a
in
st
d
e
v
ice
s
with
li
m
it
e
d
c
o
m
p
u
t
in
g
c
a
p
a
c
it
y
.
Hi
g
h
d
e
tec
ti
o
n
a
c
c
u
ra
c
y
is
c
ru
c
ial
fo
r
th
e
se
re
so
u
rc
e
-
c
o
n
stra
in
e
d
e
n
v
iro
n
m
e
n
ts,
wh
e
re
fa
lse
p
o
siti
v
e
s
c
a
n
d
isr
u
p
t
leg
it
ima
te
t
ra
ffic
a
n
d
fa
lse
n
e
g
a
ti
v
e
s
a
ll
o
w
a
tt
a
c
k
s
to
p
e
rsist.
Ho
we
v
e
r,
m
o
d
e
rn
re
i
n
fo
r
c
e
m
e
n
t
lea
rn
in
g
(RL)
a
n
d
m
a
c
h
i
n
e
lea
rn
in
g
(M
L)
in
tru
si
o
n
d
e
tec
ti
o
n
so
l
u
ti
o
n
s
o
ften
e
x
h
ib
it
p
o
o
r
g
e
n
e
ra
li
z
a
ti
o
n
d
u
e
t
o
sta
ti
c
sta
te
re
p
re
se
n
tatio
n
s.
To
a
d
d
re
ss
th
is,
th
is
p
a
p
e
r
p
r
o
p
o
se
s
a
d
e
e
p
Q
-
l
e
a
rn
in
g
(DQ
L)
fra
m
e
wo
rk
t
h
a
t
in
teg
ra
tes
K
-
m
e
a
n
s
c
l
u
ste
rin
g
d
irec
tl
y
in
to
th
e
R
L
a
c
ti
o
n
sp
a
c
e
.
U
n
li
k
e
p
ri
o
r
RL
-
b
a
se
d
IDS
m
o
d
e
ls,
o
u
r
a
p
p
r
o
a
c
h
d
y
n
a
m
ica
ll
y
in
teg
ra
tes
c
lu
ste
ri
n
g
in
t
o
th
e
lea
rn
in
g
p
ro
c
e
ss
,
e
n
a
b
li
n
g
a
d
a
p
ti
v
e
sta
te
re
p
re
se
n
tati
o
n
a
n
d
imp
ro
v
e
d
g
e
n
e
ra
li
z
a
ti
o
n
to
u
n
se
e
n
traffic
p
a
tt
e
rn
s.
T
h
e
sy
ste
m
is
fo
rm
u
late
d
a
s
a
M
a
rk
o
v
d
e
c
isio
n
p
ro
c
e
ss
wh
e
re
th
e
a
g
e
n
t
o
p
ti
m
ize
s
a
c
o
m
p
o
site
re
wa
rd
fu
n
c
ti
o
n
b
a
se
d
o
n
a
c
c
u
ra
c
y
,
p
re
c
isio
n
,
re
c
a
ll
,
a
n
d
F
1
-
sc
o
re
.
Ev
a
l
u
a
ted
o
n
th
e
N
-
Ba
Io
T
d
a
tas
e
t
u
sin
g
1
0
-
fo
l
d
c
ro
ss
-
v
a
li
d
a
ti
o
n
,
t
h
e
p
r
o
p
o
se
d
m
e
th
o
d
a
c
h
iev
e
s
a
c
las
sifica
ti
o
n
a
c
c
u
ra
c
y
o
f
9
8
.
9
5
%
a
n
d
a
we
ig
h
ted
F
1
-
sc
o
r
e
o
f
9
8
.
7
3
%
,
si
g
n
if
ica
n
tl
y
o
u
tp
e
rfo
rm
in
g
trad
it
io
n
a
l
M
L
a
n
d
RL
b
a
s
e
li
n
e
s.
Th
e
se
re
su
lt
s
d
e
m
o
n
stra
te
th
e
fra
m
e
wo
rk
'
s
e
ffe
c
ti
v
e
n
e
ss
a
s
a
sc
a
lab
le,
a
d
a
p
t
iv
e
s
o
lu
ti
o
n
f
o
r
in
tel
li
g
e
n
t
Io
T
DD
o
S
d
e
tec
ti
o
n
.
K
ey
w
o
r
d
s
:
Dee
p
Q
-
lear
n
in
g
Dis
tr
ib
u
ted
d
en
ial
o
f
s
er
v
ice
I
n
ter
n
et
o
f
th
in
g
s
s
ec
u
r
ity
Q
lear
n
in
g
R
ein
f
o
r
ce
m
en
t le
ar
n
i
n
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
:
L
an
a
Kam
la
Ah
m
ed
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scin
ec
e
an
d
I
n
f
o
r
m
atio
n
T
ec
h
n
o
lo
g
y
,
C
o
lleg
e
o
f
Scien
ce
Salah
ad
d
in
Un
iv
er
s
ity
-
E
r
b
il
Ker
k
u
k
R
o
ad
,
E
r
b
il,
I
r
aq
E
m
ail:
k
lan
am
an
atik
@
g
m
ail.
c
o
m
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
r
ap
id
ev
o
lu
tio
n
o
f
th
e
i
n
ter
n
et
o
f
th
in
g
s
(
I
o
T
)
h
as
tr
an
s
f
o
r
m
ed
h
o
w
d
ev
ices
in
te
r
ac
t
wh
ile
ex
p
o
s
in
g
s
y
s
tem
s
to
ad
v
a
n
ce
d
s
ec
u
r
ity
th
r
ea
ts
.
Su
ch
attac
k
s
,
in
clu
d
in
g
SYN
f
lo
o
d
s
an
d
UDP
am
p
lific
atio
n
,
ar
e
k
n
o
wn
as
d
is
tr
ib
u
te
d
d
e
n
i
al
-
of
-
s
er
v
ice
(
DDo
S)
attac
k
s
th
at
ex
p
lo
it
t
h
e
lim
ited
d
ef
en
s
iv
e
ca
p
ab
ilit
ies
o
f
I
o
T
h
ar
d
war
e,
g
iv
en
its
r
eso
u
r
ce
-
co
n
s
tr
ain
ed
n
atu
r
e
(
2
0
2
4
s
tatis
t
ics
s
h
o
w
th
at
o
v
er
2
1
.
3
m
illi
o
n
attac
k
s
ar
e
r
eg
is
ter
ed
)
[
1
]
.
Hig
h
d
etec
tio
n
r
ates
m
ea
n
ev
er
y
th
in
g
to
s
u
ch
r
eso
u
r
ce
-
co
n
s
tr
ain
e
d
d
ev
i
ce
s
;
f
alse
p
o
s
itiv
es
ca
n
d
is
r
u
p
t
leg
itima
te
tr
a
f
f
ic
,
b
u
t
f
alse
n
e
g
ativ
es
ca
n
en
ab
le
ad
v
er
s
ar
ial
s
u
r
v
i
v
al.
C
o
n
v
en
tio
n
al
s
ec
u
r
ity
f
r
am
ewo
r
k
s
th
at
ce
n
ter
o
n
id
e
n
tity
co
n
tr
o
l,
ac
ce
s
s
co
n
tr
o
l,
a
n
d
au
th
o
r
izatio
n
o
f
ten
ca
n
n
o
t
h
an
d
le
lar
g
e
-
s
ca
le,
d
y
n
am
ic
c
y
b
er
attac
k
s
th
at
h
av
e
b
ec
o
m
e
a
d
o
m
in
an
t
f
ea
t
u
r
e
o
f
th
e
I
o
T
[
2
]
,
[
3
]
.
B
esid
es,
h
ig
h
r
ates
o
f
m
alwa
r
e,
p
ac
k
et
-
s
n
if
f
in
g
,
an
d
s
p
o
o
f
in
g
r
eq
u
ir
e
d
etec
tio
n
m
o
d
els
th
at
ca
n
ad
ap
t
in
r
ea
l
tim
e
to
ch
an
g
in
g
tr
af
f
ic
p
atter
n
s
[
4
]
,
[
5
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
4
3
,
No
.
1
,
Ju
ly
20
2
6
:
299
-
31
3
300
Alth
o
u
g
h
p
r
im
itiv
e
m
ac
h
in
e
-
lear
n
in
g
(
ML
)
a
n
d
d
ee
p
-
lear
n
in
g
(
DL
)
m
o
d
els,
lik
e
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NNs),
r
ec
u
r
r
en
t
n
eu
r
al
n
etwo
r
k
s
(
R
NNs),
an
d
a
u
to
en
co
d
er
s
,
r
o
u
tin
el
y
m
o
n
ito
r
co
m
m
o
n
m
etr
ics
o
f
th
e
n
etwo
r
k
,
s
u
ch
as
th
e
s
o
u
r
c
e
I
P
a
d
d
r
ess
es,
th
ey
ten
d
n
o
t
to
a
d
ap
t
to
v
ar
y
in
g
tr
af
f
ic
p
atter
n
s
[
6
]
-
[
8
]
.
Key
b
o
ar
d
s
th
er
m
al
a
n
d
en
er
g
y
an
aly
s
is
with
ex
ter
io
r
m
esh
s
cr
ee
n
s
o
f
o
f
f
ice
s
p
ac
es
u
n
d
er
d
iv
e
r
s
e
clim
atic
co
n
d
itio
n
s
.
B
u
t
ex
is
t
in
g
r
ein
f
o
r
ce
m
en
t
-
b
ased
I
DS
r
ely
o
n
f
ix
ed
s
tate
m
o
d
el
lin
g
an
d
lack
ad
ap
tiv
e
f
ea
tu
r
e
ab
s
tr
ac
tio
n
[
9
]
,
[
1
0
]
.
E
v
en
th
o
u
g
h
th
e
liter
atu
r
e
in
d
icate
s
th
at
p
r
ed
ictiv
e
ac
cu
r
ac
y
h
as
im
p
r
o
v
ed
o
n
ly
m
ar
g
in
ally
,
s
ig
n
if
ican
t
ch
alle
n
g
es
r
em
ain
.
T
h
e
class
ical
ML
p
ip
elin
es
(
h
y
b
r
id
K
-
m
e
an
s
co
m
b
in
ed
with
ar
tific
ial
n
eu
r
al
n
etw
o
r
k
s
(
A
NNs)
[
1
1
]
,
s
u
p
p
o
r
t
-
v
ec
to
r
m
a
ch
in
es
[
1
2
]
,
o
r
r
an
d
o
m
-
f
o
r
est
en
s
em
b
les
[
1
3
]
)
ar
e
o
f
ten
lin
k
ed
to
eith
er
h
ig
h
co
m
p
u
tatio
n
al
co
s
t o
r
p
o
o
r
g
en
e
r
aliza
tio
n
.
Similar
ly
,
th
e
ac
cu
r
ac
y
o
f
d
ee
p
-
lea
r
n
in
g
m
o
d
els
s
u
ch
as
DE
E
P
Sh
ield
[
1
4
]
,
[
1
5
]
,
XGBo
o
s
t
[
1
6
]
,
an
d
B
i
L
STM
-
C
NN
[
1
7
]
is
also
q
u
ite
r
ea
s
o
n
ab
le,
b
u
t
th
ey
p
o
s
e
s
ig
n
if
ican
t
im
p
lem
en
tatio
n
ch
allen
g
es
in
r
eso
u
r
ce
-
lim
ited
co
n
tex
ts
[
1
8
]
,
[
1
9
]
.
At
th
e
s
am
e
tim
e,
RL
-
b
ased
m
o
d
els
lik
e
MA
DDPG
[
2
0
]
an
d
d
ee
p
Q
-
lear
n
in
g
(
DQL
)
-
I
DS
[
2
1
]
ar
e
ad
ap
ti
v
e
b
u
t
ar
e
g
en
e
r
ally
test
ed
in
s
y
n
th
etic
s
im
u
latio
n
s
an
d
h
a
v
e
n
o
t b
ee
n
p
r
o
p
er
ly
as
s
ess
ed
o
n
u
n
s
ee
n
tr
af
f
ic
s
am
p
l
es [
2
2
]
-
[
2
5
]
.
T
h
is
Stu
d
y
f
ills
th
is
g
ap
b
y
m
ak
in
g
ad
a
p
tiv
e
clu
s
ter
in
g
a
n
in
teg
r
al
p
ar
t
o
f
t
h
e
lear
n
in
g
p
r
o
ce
s
s
.
Un
lik
e
p
r
e
v
io
u
s
m
o
d
els,
we
a
ls
o
in
tr
o
d
u
ce
clu
s
ter
in
g
in
to
t
h
e
ag
e
n
t'
s
d
ec
is
io
n
s
p
ac
e,
th
e
r
eb
y
d
ev
elo
p
in
g
an
ad
ap
tiv
e
s
tate
r
e
p
r
esen
tatio
n
th
at
g
en
e
r
alize
s
to
n
ew
tr
af
f
ic
to
p
o
lo
g
ies.
I
n
p
ar
ticu
lar
,
th
e
K
-
m
ea
n
s
alg
o
r
it
h
m
is
ad
d
ed
to
th
e
DQL
ag
en
t'
s
ac
tio
n
r
ep
er
to
ir
e
t
o
en
ab
le
it to
i
d
en
tify
th
e
o
p
tim
al
n
u
m
b
er
o
f
clu
s
ter
s
(
K)
in
r
ea
l
tim
e.
As
an
ex
ten
s
io
n
o
f
th
e
n
o
n
-
a
d
ap
tiv
e
p
r
e
p
r
o
ce
s
s
in
g
,
we
u
s
e
a
co
m
p
o
s
ite
r
ewa
r
d
s
c
h
em
e
th
at
in
teg
r
ates
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
a
n
d
F1
-
s
co
r
e
t
o
jo
in
tly
o
p
tim
ize
clu
s
ter
in
g
d
ec
is
io
n
s
an
d
in
tr
u
s
io
n
-
d
etec
tio
n
p
o
licies.
T
h
is
p
ar
ad
ig
m
e
n
ab
les
th
e
ag
en
t
to
o
b
tain
a
h
ig
h
er
-
lev
el
s
tate
ab
s
tr
ac
tio
n
an
d
s
tab
ilize
Q
-
v
alu
e
co
n
v
er
g
en
ce
,
th
e
r
eb
y
im
p
r
o
v
i
n
g
th
e
ef
f
ec
tiv
en
ess
o
f
DDo
S
d
etec
tio
n
.
T
h
e
m
ain
co
n
tr
i
b
u
tio
n
s
o
f
t
h
e
wo
r
k
ar
e
s
u
m
m
ar
ize
d
as
f
o
llo
ws:
th
e
cr
ea
tio
n
o
f
a
DQL
ar
ch
itectu
r
e
th
at
allo
ws
cr
ea
tin
g
ad
ap
tiv
e
r
ep
r
esen
tatio
n
s
b
ased
o
n
th
e
ac
tio
n
s
p
ac
e,
th
e
d
esig
n
o
f
a
n
ew
co
m
p
o
s
ite
m
ec
h
an
is
m
o
f
r
ewa
r
d
s
t
o
o
p
t
im
ize
th
em
s
im
u
ltan
eo
u
s
ly
,
a
n
d
an
em
p
ir
ical
v
alid
atio
n
is
co
n
d
u
cted
with
a
s
ig
n
if
ican
t
am
o
u
n
t
o
f
r
ig
o
r
u
s
in
g
ten
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
o
n
th
e
N
-
B
aI
o
T
d
ataset
to
p
r
o
v
e
th
e
s
u
p
er
io
r
ity
o
v
er
tr
ad
itio
n
al
b
aselin
es.
T
h
e
r
est
o
f
th
e
m
an
u
s
cr
ip
t
will
b
e
s
tr
u
ctu
r
ed
an
d
p
r
esen
ted
as
f
o
llo
ws:
s
ec
tio
n
2
will
d
escr
ib
e
th
e
m
ater
ials
an
d
m
eth
o
d
o
lo
g
y
,
s
ec
tio
n
3
will
p
r
esen
t
th
e
em
p
ir
ical
f
in
d
in
g
s
,
an
d
s
ec
tio
n
4
will
p
r
esen
t th
e
co
n
clu
s
io
n
an
d
r
ec
o
m
m
en
d
atio
n
s
f
o
r
f
u
tu
r
e
r
esear
ch
.
2.
M
E
T
H
O
D
T
o
m
ain
tain
tr
an
s
p
a
r
en
cy
,
r
ep
r
o
d
u
cib
ilit
y
,
an
d
s
cien
tific
r
i
g
o
r
,
th
e
s
ec
tio
n
d
escr
ib
es
th
e
in
s
tr
u
m
en
ts
,
m
ater
ials
,
an
d
ex
p
er
im
en
tal
m
eth
o
d
s
ap
p
lied
wh
e
n
ca
r
r
y
i
n
g
o
u
t
th
e
cu
r
r
en
t
s
tu
d
y
.
Fig
u
r
e
1
illu
s
tr
ates
th
e
ex
p
er
im
en
tal
p
ip
elin
e
u
s
ed
i
n
th
e
p
r
esen
t
r
esear
ch
.
I
t
s
ta
r
ts
with
th
e
a
b
s
o
r
p
tio
n
o
f
th
e
N
-
B
aI
o
T
d
ataset,
co
n
tin
u
es
with
d
ata
p
r
ep
r
o
ce
s
s
in
g
an
d
f
ea
tu
r
e
en
g
in
ee
r
i
n
g
,
an
d
en
d
s
with
th
e
cr
ea
tio
n
an
d
test
in
g
o
f
two
r
ein
f
o
r
ce
m
e
n
t
lear
n
in
g
m
o
d
els,
i.e
.
,
Q
-
lear
n
in
g
a
n
d
DQL
.
T
h
e
two
m
o
d
els
co
m
b
in
e
th
e
K
-
m
ea
n
s
clu
s
ter
in
g
m
eth
o
d
a
n
d
ar
e
tr
ain
ed
an
d
ev
alu
ated
u
s
in
g
K
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
.
T
h
e
p
ip
elin
e
co
n
clu
d
es
with
a
co
m
p
ar
at
iv
e
p
er
f
o
r
m
an
ce
a
n
al
y
s
is
u
s
in
g
p
r
ed
ef
in
e
d
ev
alu
ati
o
n
m
etr
ics.
Fig
u
r
e
1
.
E
x
p
er
im
e
n
tal
p
ip
elin
e
in
teg
r
atin
g
a
d
ap
tiv
e
clu
s
ter
i
n
g
with
DQL
2
.
1
.
Da
t
a
s
et
d
escript
io
n
T
h
e
d
ataset
u
s
ed
to
tr
ain
th
e
p
r
o
p
o
s
ed
m
o
d
el
is
th
e
NB
aI
o
T
d
ataset,
p
u
b
licly
av
ailab
le
o
n
Kag
g
le
[
2
6
]
.
T
h
is
d
ataset
co
m
p
r
is
es
n
etwo
r
k
tr
af
f
ic
f
r
o
m
n
in
e
r
ea
l
-
wo
r
ld
I
o
T
d
ev
ices,
s
u
ch
as
we
b
ca
m
s
,
th
er
m
o
s
tats
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
Dee
p
Q
lea
r
n
in
g
a
l
g
o
r
ith
m
fo
r
d
etec
tin
g
DDo
S
a
tta
ck
s
o
n
I
o
T d
ev
ices
(
La
n
a
K
a
mla
A
h
m
ed
)
301
an
d
b
ab
y
m
o
n
ito
r
s
,
ca
p
tu
r
ed
d
u
r
in
g
r
eg
u
lar
o
p
er
ati
o
n
a
n
d
u
n
d
er
a
v
a
r
iety
o
f
attac
k
co
n
d
itio
n
s
,
in
clu
d
in
g
SYN
f
lo
o
d
,
UDP
f
lo
o
d
,
an
d
d
ata
ex
f
iltra
tio
n
.
Alth
o
u
g
h
th
e
d
ataset
is
r
ea
li
s
tic
in
ter
m
s
o
f
th
e
d
y
n
am
ics
o
f
a
ty
p
ical
I
o
T
d
ev
ice,
it
h
as
a
d
is
p
r
o
p
o
r
tio
n
ate
class
d
is
tr
ib
u
t
io
n
,
with
th
e
attac
k
tr
af
f
ic
s
a
m
p
le
s
ig
n
if
ican
tly
lar
g
er
th
an
t
h
at
o
f
n
o
r
m
al
tr
a
f
f
ic.
T
h
is
im
b
alan
ce
m
ay
h
a
v
e
an
im
p
ac
t
o
n
th
e
m
o
d
el
tr
a
in
in
g
a
n
d
o
u
tp
u
ts
.
T
ab
le
1
will
b
e
th
e
s
u
m
m
a
r
y
o
f
th
e
k
ey
p
o
in
ts
r
elate
d
to
t
h
e
d
ataset,
s
u
ch
as
s
am
p
le
s
izes
an
d
d
ataset
an
d
tr
af
f
ic
f
ea
tu
r
e
s
izes [
2
6
]
.
T
ab
le
1
.
C
h
ar
ac
ter
is
tics
an
d
s
am
p
le
s
tr
u
ctu
r
e
(
N
-
B
aI
o
T
d
ataset)
A
t
t
r
i
b
u
t
e
D
e
scri
p
t
i
o
n
N
u
mb
e
r
o
f
sam
p
l
e
s
7
,
0
0
0
,
0
0
0
+
N
u
mb
e
r
o
f
f
e
a
t
u
r
e
s
1
1
5
Tr
a
f
f
i
c
t
y
p
e
s
B
e
n
i
g
n
,
M
i
r
a
i
-
b
a
se
d
,
B
A
S
H
LI
TE
-
b
a
s
e
d
,
T
C
P
,
U
D
P
f
l
o
o
d
s
,
s
c
a
n
s,
c
o
mb
o
s
Ty
p
e
o
f
b
o
t
n
e
t
M
i
r
a
i
a
n
d
B
A
S
H
LI
TE
(
a
.
k
.
a
.
G
a
f
g
y
t
)
C
l
a
s
si
f
i
c
a
t
i
o
n
f
r
a
mi
n
g
B
i
n
a
r
y
(
b
e
n
i
g
n
v
s
a
t
t
a
c
k
)
a
n
d
mu
l
t
i
c
l
a
ss (1
1
t
o
t
a
l
c
l
a
s
ses)
K
e
y
f
e
a
t
u
r
e
d
o
ma
i
n
s
P
r
o
t
o
c
o
l
s
t
a
t
s,
p
a
c
k
e
t
r
a
t
e
,
I
C
M
P
f
r
e
q
u
e
n
c
y
,
st
r
e
a
m a
n
o
ma
l
i
e
s
C
l
a
s
s
d
i
s
t
r
i
b
u
t
i
o
n
I
mb
a
l
a
n
c
e
d
-
b
e
n
i
g
n
d
o
mi
n
a
t
e
s
I
o
T
d
e
v
i
c
e
s
t
y
p
e
s
C
o
l
l
e
c
t
e
d
f
r
o
m
d
i
f
f
e
r
e
n
t
t
y
p
e
s
o
f
I
o
T
r
e
a
l
d
e
v
i
c
e
i
n
c
l
u
d
i
n
g
Th
e
r
mo
s
t
a
t
,
P
h
i
l
i
p
s
B
a
b
y
mo
n
i
t
o
r
,
p
r
o
v
i
s
i
o
n
sec
u
r
i
t
y
c
a
m
e
r
a
,
s
a
ms
u
n
g
sm
a
r
t
TV
,
S
i
mp
l
e
H
o
m
e
C
a
mer
a
s,
a
n
d
W
e
M
o
S
mart
P
l
u
g
,
D
a
mi
n
i
D
o
o
r
b
e
l
l
,
Ec
o
B
e
e
T
o
.
2
.
2
.
Da
t
a
p
re
pro
ce
s
s
ing
T
h
e
m
o
s
t im
p
o
r
tan
t p
a
r
t in
th
e
p
r
ep
ar
atio
n
o
f
th
e
d
atasets
was p
r
ep
r
o
ce
s
s
in
g
.
T
h
e
in
itial p
h
ase
o
f
th
e
p
r
o
ce
s
s
was
to
f
ill
g
ap
s
in
v
alu
es
with
th
e
h
elp
o
f
th
e
n
p
.
n
an
to
n
u
m
f
u
n
ctio
n
,
wh
ic
h
s
u
b
s
titu
tes
m
is
s
in
g
v
alu
es
with
n
u
m
er
ical
esti
m
at
es.
T
h
e
a
d
d
itio
n
al
d
im
e
n
s
io
n
a
lity
r
ed
u
ctio
n
was
d
o
n
e
th
r
o
u
g
h
th
e
r
e
m
o
v
al
o
f
r
ed
u
n
d
an
t
lo
w
-
v
ar
ian
ce
an
d
D
Do
S
-
d
etec
tio
n
-
ir
r
elev
a
n
t
f
ea
tu
r
es,
wh
ich
h
el
p
ed
t
o
f
ilter
t
h
e
n
u
m
b
er
o
f
f
ac
to
r
s
th
at
m
ay
n
o
t
co
n
tr
ib
u
te
to
th
e
tr
ain
in
g
o
f
t
h
e
m
o
d
els.
Su
b
s
eq
u
en
t
n
o
r
m
al
izatio
n
o
f
co
n
ti
n
u
o
u
s
f
ea
t
u
r
es
was
p
er
f
o
r
m
ed
u
s
in
g
Z
-
s
co
r
e
s
ta
n
d
ar
d
izatio
n
with
Stan
d
ar
d
S
ca
ler
,
wh
ich
s
tab
ilizes
th
e
d
ata
d
is
tr
ib
u
tio
n
an
d
im
p
r
o
v
es
lear
n
i
n
g
[
2
7
]
.
A
n
o
th
er
ch
an
g
e
is
th
at
th
e
s
cik
it
-
lear
n
KB
in
s
Dis
cr
etize
r
was
u
s
ed
to
g
en
e
r
ate
s
tatis
t
ically
e
q
u
alize
d
b
in
b
o
u
n
d
ar
ies,
th
er
e
b
y
p
r
o
v
i
d
in
g
a
d
ee
p
er
s
tate
r
ep
r
esen
tatio
n
f
o
r
th
e
DQL
ag
en
t.
Sp
ec
if
ically
,
th
e
N
-
B
aI
o
T
d
a
taset
is
n
o
t
ca
teg
o
r
ical;
h
o
w
ev
er
,
s
in
ce
t
h
e
r
aw
i
n
p
u
t
was
d
is
cr
etize
d
,
th
e
r
esu
ltin
g
b
in
s
wer
e
ca
teg
o
r
ica
l
an
d
co
n
v
e
r
ted
in
to
a
n
en
u
m
er
ated
,
o
n
e
-
h
o
t
f
o
r
m
at
to
m
a
k
e
th
em
co
m
p
atib
le
with
th
e
u
n
d
e
r
ly
in
g
lea
r
n
in
g
m
o
d
el.
2
.
3
.
M
o
del
a
rc
hite
ct
ure
T
h
e
cu
r
r
en
t
r
esear
ch
o
u
tlin
es
th
e
f
r
am
ewo
r
k
f
o
r
a
DQL
s
y
s
tem
im
p
lem
en
ted
as
a
f
u
lly
co
n
n
ec
te
d
f
ee
d
-
f
o
r
war
d
n
eu
r
al
n
etwo
r
k
,
as
s
h
o
wn
in
Fig
u
r
e
2
.
T
h
e
n
etwo
r
k
h
as
a
n
etwo
r
k
in
p
u
t
lay
er
,
two
h
id
d
en
lay
er
s
,
an
d
an
o
u
tp
u
t
lay
er
.
No
n
lin
ea
r
p
atter
n
s
in
th
e
in
p
u
t
d
ata
ar
e
ca
p
tu
r
ed
b
y
th
e
h
i
d
d
en
lay
er
s
u
s
in
g
th
e
r
ec
tifie
d
lin
ea
r
u
n
it
(
R
eL
U)
ac
tiv
atio
n
f
u
n
ctio
n
,
wh
ich
h
as
b
ee
n
p
r
o
v
en
e
f
f
ec
tiv
e
in
d
ee
p
lear
n
i
n
g
a
r
ch
itectu
r
es
[
2
8
]
.
Fu
r
th
er
m
o
r
e,
to
a
d
d
r
ess
class
im
b
alan
ce
an
d
en
s
u
r
e
h
ig
h
d
etec
tio
n
q
u
ality
[
2
9
]
,
we
d
esig
n
ed
a
n
o
v
el
c
o
m
p
o
s
ite
r
e
war
d
f
u
n
ctio
n
.
Ou
r
r
ewa
r
d
s
ig
n
al,
u
n
lik
e
s
tan
d
a
r
d
m
o
d
els,
c
o
m
b
in
es
Acc
u
r
ac
y
,
Pre
cisi
o
n
,
R
ec
all,
an
d
F1
-
s
co
r
e
(
R
=
0
.
2
5
*
Acc
+
0
.
2
5
*
Pre
c
+
0
.
2
5
*
R
ec
+
0
.
2
5
*
F1
)
.
Alth
o
u
g
h
th
e
r
ewa
r
d
f
u
n
ctio
n
is
co
m
p
u
ted
f
r
o
m
th
e
g
r
o
u
n
d
-
tr
u
th
la
b
els
o
f
th
e
N
-
B
aI
o
T
d
ataset,
th
is
is
a
tactica
l
d
esig
n
d
ec
is
io
n
f
o
r
th
e
o
f
f
lin
e
tr
ain
i
n
g
s
tag
e.
T
h
is
en
ab
les
th
e
ag
en
t
to
lear
n
an
o
p
tim
al
class
if
ica
t
io
n
p
o
licy
th
at
ca
n
u
ltima
tely
b
e
ap
p
lied
i
n
an
au
to
n
o
m
o
u
s
,
lab
el
-
f
r
ee
en
v
ir
o
n
m
en
t.
A
m
u
ltid
im
en
s
io
n
al
f
ee
d
b
ac
k
allo
ws
r
ed
u
cin
g
th
e
r
ate
o
f
f
alse
p
o
s
itiv
es
an
d
f
alse
n
eg
ativ
es
at
th
e
s
am
e
tim
e,
wh
ich
is
n
ec
es
s
a
r
y
to
g
u
ar
a
n
tee
th
e
av
ailab
ilit
y
o
f
s
er
v
ices
i
n
r
eso
u
r
ce
-
lim
ited
I
o
T
s
ettin
g
s
.
T
h
e
o
u
tp
u
t
lay
e
r
u
s
es
a
lin
ea
r
ac
ti
v
atio
n
t
o
esti
m
ated
Q
-
v
alu
es
o
f
ev
er
y
p
o
s
s
ib
le
ac
tio
n
av
ailab
le,
th
er
ef
o
r
e
allo
win
g
th
e
ag
en
t
to
ca
teg
o
r
ize
th
e
n
etwo
r
k
tr
af
f
ic
as
b
en
ig
n
o
r
m
alicio
u
s
ac
c
o
r
d
in
g
to
d
y
n
am
ically
esti
m
ated
cl
u
s
ter
ass
ig
n
m
en
ts
.
I
n
o
r
d
e
r
to
m
a
k
e
th
e
r
es
p
o
n
s
e
to
c
h
a
n
g
i
n
g
t
r
a
f
f
ic
c
o
n
f
ig
u
r
a
t
io
n
s
o
f
t
h
e
I
o
T
n
e
tw
o
r
k
s
m
o
r
e
f
le
x
i
b
l
e,
K
-
m
ea
n
s
cl
u
s
t
er
in
g
is
u
s
e
d
to
co
m
p
le
m
e
n
t
t
h
e
DQ
L
f
r
a
m
e
w
o
r
k
.
T
h
e
p
r
o
c
ess
o
f
in
te
g
r
ati
o
n
al
lo
ws
s
t
at
es o
f
t
h
e
n
et
wo
r
k
s
t
o
b
e
ag
g
r
e
g
a
te
d
i
n
a
s
tr
u
ct
u
r
ed
f
o
r
m
to
g
i
v
e
s
ta
te
i
n
f
o
r
m
at
io
n
o
f
t
h
e
h
i
g
h
-
d
i
m
e
n
s
i
o
n
al
s
tat
es
o
f
t
h
e
s
y
s
te
m
.
T
u
p
l
es o
f
(
s
t
ate
,
ac
ti
o
n
,
r
ew
ar
d
,
n
e
x
t
s
t
ate
)
tr
a
n
s
iti
o
n
s
o
f
s
t
at
e
ar
e
f
e
d
i
n
t
o
a
r
e
p
l
ay
b
u
f
f
er
i
n
tr
ai
n
i
n
g
t
o
r
e
g
u
la
r
i
ze
th
e
tr
ai
n
i
n
g
p
r
o
c
e
s
s
a
n
d
t
o
r
e
d
u
c
e
c
o
r
r
el
at
i
o
n
s
am
o
n
g
s
t
r
e
lat
ed
s
a
m
p
les
.
I
t
ap
p
lies
ε
-
g
r
e
e
d
y
alg
o
r
it
h
m
i
n
o
r
d
e
r
to
m
ak
e
a
t
r
ad
e
-
o
f
f
wi
th
e
x
p
l
o
r
ati
o
n
an
d
e
x
p
l
o
ita
ti
o
n
wi
th
t
h
e
ai
m
o
f
m
a
x
i
m
iz
in
g
lo
n
g
-
te
r
m
cu
m
u
lat
iv
e
r
e
tu
r
n
s
.
O
v
e
r
al
l,
t
h
is
ar
c
h
it
ec
t
u
r
e
will
al
lo
w
D
QL
a
g
en
t
t
o
b
e
t
r
ai
n
ed
t
o
a
d
o
p
t
u
s
ef
u
l
p
o
l
ici
es
t
o
d
et
ec
t
DD
o
S
i
n
I
o
T
e
n
v
ir
o
n
m
en
ts
.
T
h
e
a
r
c
h
it
ec
t
u
r
e
p
r
o
v
i
d
e
s
a
s
tat
e
c
h
a
n
g
e
s
eq
u
e
n
ce
,
a
n
d
als
o
u
s
es
ad
a
p
ti
v
e
clu
s
te
r
,
a
n
d
th
is
m
a
k
es
i
t
p
o
s
s
i
b
le
t
o
le
ar
n
a
n
d
d
e
te
ct
p
atte
r
n
s
i
n
h
i
g
h
-
d
i
m
e
n
s
i
o
n
a
l
n
e
tw
o
r
k
t
r
a
f
f
ic
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
4
3
,
No
.
1
,
Ju
ly
20
2
6
:
299
-
31
3
302
Fig
u
r
e
2
.
Pro
p
o
s
ed
DQL
ar
c
h
i
tectu
r
e
f
o
r
d
y
n
a
m
ic
Q
-
v
alu
e
a
p
p
r
o
x
im
atio
n
2
.
4
.
P
r
o
po
s
ed
enha
nced
DQ
L
det
ec
t
io
n m
o
del
T
o
s
elec
t
an
d
ev
alu
ate
a
r
esea
r
ch
m
o
d
el
o
f
DQL
,
th
e
p
r
esen
t
p
ap
er
f
o
llo
wed
a
s
y
s
tem
atic
ap
p
r
o
ac
h
an
d
b
eg
a
n
b
y
estab
lis
h
in
g
t
h
e
m
ain
h
y
p
e
r
p
ar
am
ete
r
s
,
wh
ich
d
ef
in
ed
t
h
e
b
ac
k
g
r
o
u
n
d
o
f
th
e
ex
p
er
im
e
n
t
p
r
o
to
co
l,
i.e
.
,
th
e
d
ataset,
lea
r
n
in
g
r
ate,
d
is
co
u
n
t
f
ac
to
r
,
m
em
o
r
y
s
ize
an
d
th
e
n
u
m
b
er
o
f
K
-
f
o
l
d
s
.
T
h
e
N
-
B
aI
o
T
was
in
th
e
f
ir
s
t
s
tep
s
p
r
ep
r
o
ce
s
s
ed
with
th
e
n
o
r
m
aliza
tio
n
an
d
f
ea
tu
r
e
e
n
co
d
in
g
to
e
n
h
an
ce
th
e
q
u
ality
o
f
d
ata
an
d
co
r
r
ec
t
th
e
p
ar
tia
l
m
is
s
in
g
n
ess
.
Fo
llo
win
g
th
is
,
th
e
K
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
d
iv
id
ed
th
e
d
ata
to
f
o
r
m
tr
ain
i
n
g
an
d
cr
o
s
s
-
v
ali
d
atio
n
f
o
l
d
s
,
th
u
s
allo
win
g
i
n
ten
s
iv
e
ass
ess
m
en
t
o
f
s
ev
er
al
f
o
ld
s
[
3
0
]
.
T
h
is
wo
r
k
f
lo
w
m
ain
tain
s
p
ar
am
ete
r
s
ettin
g
s
in
a
p
r
o
ce
s
s
a
s
s
h
o
wn
in
Fig
u
r
e
3
.
I
t
h
elp
s
in
th
e
u
ltima
te
ev
alu
atio
n
o
f
ag
g
r
eg
ate
p
e
r
f
o
r
m
an
ce
m
ea
s
u
r
es a
n
d
,
h
e
n
ce
,
g
o
o
d
m
o
n
ito
r
in
g
o
f
DDo
S a
ttack
s
.
On
to
p
o
f
th
is
wo
r
k
f
lo
w,
th
e
s
y
s
tem
d
ir
ec
tly
in
clu
d
es
a
K
-
m
ea
n
s
clu
s
ter
in
g
m
o
d
u
le,
th
e
DQL
ag
en
t,
a
m
eta
-
lear
n
er
[
3
1
]
.
T
h
e
ag
e
n
t,
as
o
p
p
o
s
ed
to
u
s
in
g
th
e
s
tatic
m
eth
o
d
o
f
clu
s
ter
in
g
,
f
ig
u
r
es
o
u
t
th
e
o
p
tim
al
n
u
m
b
er
o
f
clu
s
ter
s
(
K)
o
n
th
e
f
ly
,
an
d
th
is
is
co
n
s
id
er
ed
to
b
e
th
e
ac
tio
n
p
er
f
o
r
m
ed
b
y
th
e
K
-
m
ea
n
s
m
o
d
u
le.
T
h
e
en
a
b
lin
g
a
d
ap
tiv
e
ap
p
r
o
ac
h
e
n
ab
les
n
o
is
y
f
ea
tu
r
es
to
b
e
clu
s
ter
ed
au
to
m
atica
lly
an
d
b
o
o
s
ts
th
e
s
ep
ar
atio
n
o
f
class
es
in
c
o
m
p
l
icate
d
tr
af
f
ic.
T
h
e
r
e
p
lay
m
e
m
o
r
y
(
D)
is
em
p
lo
y
ed
to
s
tab
ilize
th
e
n
etwo
r
k
b
y
k
ee
p
in
g
p
r
ev
io
u
s
ex
p
er
ien
ce
s
,
an
d
an
e
p
s
ilo
n
-
g
r
ee
d
y
p
o
licy
is
em
p
lo
y
ed
to
m
ak
e
th
e
p
r
o
g
r
am
b
ala
n
ce
d
in
ex
p
lo
r
in
g
n
ew
attac
k
p
atter
n
s
in
ad
d
itio
n
to
th
e
p
r
ev
i
o
u
s
o
n
es.
T
h
e
wh
o
le
r
atio
n
ale
o
f
th
e
wo
r
k
in
g
m
ec
h
an
is
m
o
f
th
e
s
y
s
tem
,
t
h
e
p
a
r
ticu
lar
p
ar
am
eter
s
ettin
g
,
a
n
d
th
e
d
esig
n
elem
e
n
ts
ar
e
s
u
m
m
ar
ized
in
T
ab
le
1
[
3
2
]
a
n
d
f
o
r
m
alize
d
in
Alg
o
r
ith
m
1
.
Fig
u
r
e
3
.
Sy
s
tem
wo
r
k
f
lo
w
f
o
r
ag
en
t
-
b
ased
o
p
tim
al
clu
s
ter
s
elec
tio
n
T
ab
le
2
.
DQL
fr
am
ewo
r
k
p
ar
a
m
eter
s
an
d
d
esig
n
elem
e
n
ts
s
u
m
m
ar
y
P
a
r
a
me
t
e
r
D
e
f
i
n
i
t
i
o
n
S
t
a
t
e
(
S
)
D
i
scret
e
n
e
t
w
o
r
k
t
r
a
f
f
i
c
c
h
a
r
a
c
t
e
r
i
st
i
c
s o
f
t
h
e
d
a
t
a
se
t
A
c
t
i
o
n
(
A
)
C
h
o
o
si
n
g
t
h
e
b
e
st
n
u
m
b
e
r
o
f
c
l
u
st
e
r
s
(
K
)
o
f
t
h
e
K
-
m
e
a
n
s m
o
d
u
l
e
.
R
e
w
a
r
d
(
R
)
A
c
o
m
p
o
si
t
e
si
g
n
a
l
:
0
.
2
5
*
(
A
c
c
u
r
a
c
y
+
P
r
e
c
i
si
o
n
+
R
e
c
a
l
l
+
F
1
-
sc
o
r
e
)
P
o
l
i
c
y
A
n
e
p
s
i
l
o
n
-
g
r
e
e
d
y
s
t
r
a
t
e
g
y
(
st
a
r
t
i
n
g
a
t
1
.
0
,
d
e
c
a
y
i
n
g
t
o
0
.
0
1
)
Q
(
st
,
a
t
)
C
u
r
r
e
n
t
Q
-
v
a
l
u
e
s
t
a
t
e
a
n
d
a
c
t
i
o
n
α (Lea
r
n
i
n
g
r
a
t
e
)
R
e
g
u
l
a
t
i
n
g
t
h
e
w
e
i
g
h
t
o
f
n
e
w
i
n
f
o
r
m
a
t
i
o
n
γ
(
D
i
sc
o
u
n
t
f
a
c
t
o
r
)
D
e
t
e
r
m
i
n
i
n
g
t
h
e
i
m
p
o
r
t
a
n
c
e
o
f
f
u
t
u
r
e
r
e
w
a
r
d
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
Dee
p
Q
lea
r
n
in
g
a
l
g
o
r
ith
m
fo
r
d
etec
tin
g
DDo
S
a
tta
ck
s
o
n
I
o
T d
ev
ices
(
La
n
a
K
a
mla
A
h
m
ed
)
303
Alg
o
r
ith
m
1
.
E
n
h
a
n
ce
d
DQL
f
o
r
I
o
T
DDo
S
d
etec
tio
n
in
pu
t:
da
ta
se
t_
fo
ld
er
,
al
ph
a
(α
),
ga
mm
a
(γ
),
ϵ_
st
ar
t,
ϵ_
de
ca
y,
ϵ_
m
in
,n
um
_e
pi
so
de
s,
replaymemory d, clustersize (possible
-
ks),k_fold
output:
performance metrics, confusion matrix
1. Initialize replay memory
2. initialize q
-
network q(s, a; θ) with random weights
3. set ϵ = ϵ_start
4. for each file in dataset_folder do
4.1 Preprocess x (normalize features, handle missing values
)
4.2 Perform k
-
fold cross
-
validation:
for each fold in k_fold do
Split data into xtrain, ytrain, xtest, ytest
for episode = 1 to num_episodes do
4.2.1 set state = get_features(xtrain)
4.2.2 while training_state
exists do
if (random(0,1) < ϵ) then
set action = randomaction()
else
set action = argmax(q(state))
end if
4.2.2.
1 Perform k
-
means clustering using k = action
4.2.2.2 Map clusters to labels using majority voting
4.2.2.3 set acc = calc_accuracy()
set prec = calc_precision()
set rec = calc_recal
l()
set f1 = calc_f1_score()
4.2.2.4 set reward = (0.25 * acc) + (0.25 * prec) + (0.25 * rec) + (0.25 * f1)
4.2.2.5 store (state, action, reward, next_state) in d
4.2.2.6 update q(state, action) using bellm
an:
set q = q + α * (reward + γ * max_q
-
q)
4.2.2.7 set ϵ = max(ϵ * ϵ_decay, ϵ_min)
end while
end for
4.3 set optimal_action = argmax(q(test_state))
4.4 Evaluate xtest using k
-
means with k = optimal_action
4.5 Calculate final metrics and confusion matrix
end for
END FOR
5. AGGREGATE metrics_across_folds.
6. RETURN overall metrics, Confusion Matrix.
T
h
e
DQL
m
o
d
el
h
y
p
er
p
a
r
am
e
ter
s
wer
e
ch
o
s
en
to
g
u
a
r
an
tee
s
tab
le
co
n
v
er
g
en
ce
a
n
d
a
s
tr
o
n
g
ab
ilit
y
to
d
etec
t
d
if
f
e
r
en
t
p
atter
n
s
o
f
attac
k
s
.
T
h
e
lear
n
in
g
r
ate
(
0
.
0
0
1
)
was
s
elec
ted
to
h
av
e
a
b
alan
ce
b
etwe
en
th
e
s
p
ee
d
at
wh
ich
th
e
weig
h
ts
ar
e
u
p
d
ated
;
th
e
h
ig
h
er
t
h
e
v
alu
es,
t
h
e
f
aster
th
e
n
etwo
r
k
wo
u
ld
g
et
awa
y
,
an
d
t
h
e
lo
wer
th
e
v
alu
es,
t
h
e
s
o
o
n
e
r
t
h
e
n
etwo
r
k
wo
u
ld
lear
n
.
T
h
e
d
is
co
u
n
t
f
ac
to
r
(
γ
)
was
d
eter
m
in
ed
to
b
e
0
.
9
5
s
o
th
at
th
e
ag
en
t
co
u
ld
m
o
d
el
th
e
f
u
tu
r
e
n
etwo
r
k
s
tates,
an
d
th
is
is
n
ec
ess
ar
y
in
o
r
d
er
to
m
o
d
el
t
h
e
tim
e
d
ep
en
d
e
n
cies
o
f
a
m
u
lti
-
s
tag
e
DDo
S
attac
k
.
T
h
e
ep
s
ilo
n
-
d
e
ca
y
s
ch
ed
u
le
b
eg
in
s
with
1
.
0
an
d
d
ec
ay
s
to
0
.
0
1
th
r
o
u
g
h
o
u
t
th
e
f
ir
s
t
1
0
0
ep
is
o
d
es,
wh
ich
allo
ws
th
e
a
g
en
t
to
ex
p
lo
r
e
th
e
h
ig
h
-
d
im
en
s
i
o
n
al
f
ea
tu
r
es
o
f
N
-
B
aI
o
T
b
r
o
a
d
ly
at
th
e
s
tar
t
b
ef
o
r
e
ch
a
n
g
in
g
o
n
to
a
s
tab
le
e
x
p
lo
itatio
n
p
o
licy
.
T
h
e
s
elec
ted
v
alu
es
c
o
r
r
esp
o
n
d
to
th
e
b
est
v
alid
atio
n
r
esu
lt
s
o
b
tain
ed
d
u
r
in
g
th
e
e
x
p
er
i
m
en
tal
ev
alu
atio
n
.
Sp
ec
if
ical
ly
,
α
c
o
n
tr
o
ls
th
e
m
ag
n
itu
d
e
o
f
weig
h
t
u
p
d
ates
in
Q
-
n
etwo
r
k
o
p
tim
izatio
n
,
t
h
er
eb
y
en
s
u
r
in
g
p
r
u
n
e
d
lear
n
i
n
g
o
f
k
n
o
wled
g
e
.
γ
,
o
n
th
e
o
th
er
h
an
d
,
h
ig
h
lig
h
ts
th
e
f
u
tu
r
e
b
en
ef
its
,
allo
win
g
th
e
ag
e
n
t
to
ta
k
e
in
to
co
n
s
i
d
er
atio
n
l
o
n
g
-
te
r
m
b
en
ef
its
in
th
e
co
m
b
in
atio
n
o
f
cu
m
u
lativ
e
r
ewa
r
d
s
.
T
h
e
ex
p
l
o
r
atio
n
co
e
f
f
icien
t
(
ϵ
)
is
th
e
ten
s
io
n
b
etwe
en
th
e
d
esire
s
to
ex
p
lo
r
e
an
d
to
ex
p
lo
it
in
f
av
o
r
o
f
an
ac
tio
n
s
el
ec
tio
n
p
r
o
ce
s
s
th
at
n
av
ig
ates
th
e
ag
en
t
b
etwe
en
p
o
licy
r
ef
in
em
en
t
an
d
s
u
cc
ess
iv
e
iter
atio
n
.
T
h
is
is
s
to
r
ed
in
a
r
ep
lay
b
u
f
f
er
(
D)
to
r
ed
u
ce
t
r
ain
in
g
in
s
tab
ilit
y
.
E
p
is
o
d
ic
ex
p
e
r
ien
ce
s
,
in
clu
d
in
g
ac
tio
n
s
,
s
tates,
an
d
r
e
war
d
s
,
ar
e
tem
p
o
r
ar
ily
s
to
r
ed
h
er
e,
allo
win
g
m
in
ib
atch
es
to
b
e
s
to
ch
asti
ca
lly
ex
tr
ac
ted
a
n
d
r
e
d
u
cin
g
co
r
r
elatio
n
s
in
h
er
en
t
in
tem
p
o
r
ally
o
r
d
e
r
ed
d
ata.
T
h
e
s
ize
o
f
th
e
ch
o
s
en
b
atc
h
d
eter
m
in
es
h
o
w
m
an
y
e
x
p
er
ie
n
ce
s
ar
e
ad
d
ed
t
o
th
e
b
u
f
f
er
ea
c
h
tim
e
an
o
p
tim
i
za
tio
n
s
tep
is
p
er
f
o
r
m
ed
,
wh
ich
,
in
tu
r
n
,
in
f
lu
e
n
ce
s
th
e
co
n
v
er
g
en
ce
r
ate
an
d
o
v
er
all
lear
n
in
g
s
tab
ilit
y
.
B
esid
es,
th
e
ε
-
d
ec
ay
s
ch
e
d
u
le
a
n
d
m
in
im
al
ex
p
lo
r
atio
n
t
h
r
esh
o
ld
(
θ)
d
e
f
in
e
th
e
tem
p
o
r
al
d
ec
a
d
en
ce
o
f
ε
an
d
p
lace
a
n
ab
s
o
lu
te
lo
wer
lim
it,
e
n
s
u
r
in
g
th
at
o
cc
asio
n
al
e
x
p
lo
r
ato
r
y
b
eh
av
io
r
is
m
ai
n
tain
ed
d
u
r
in
g
t
h
e
tr
ain
in
g
h
is
to
r
y
.
I
t
u
s
es
th
e
Q
-
n
etwo
r
k
,
wh
ich
i
s
r
ef
er
r
ed
to
as
Q
(
s
,
a,
θ)
,
to
ap
p
r
o
x
im
ate
Q
-
v
alu
es
o
f
ac
tio
n
-
s
tate
p
air
s
,
an
d
θ
is
th
e
tr
ain
ab
le
p
ar
am
eter
o
f
t
h
e
n
eu
r
al
n
etwo
r
k
.
T
h
e
n
u
m
b
e
r
o
f
t
r
ain
in
g
e
p
is
o
d
es (
Nu
m
ep
ep
is
o
d
es)
tells
th
e
am
o
u
n
t o
f
tr
ai
n
in
g
.
T
h
e
p
h
ase
o
f
esti
m
atio
n
o
f
th
e
K
-
m
ea
n
s
clu
s
ter
in
g
is
co
n
tin
g
en
t o
n
th
e
v
ar
iety
o
f
p
o
s
s
ib
le
ca
n
d
id
ate
clu
s
ter
v
al
u
es
(
KS)
.
I
t
g
iv
es
th
e
a
g
en
t
g
u
id
elin
e
s
o
n
th
e
way
to
d
eter
m
in
e
t
h
e
b
est
n
u
m
b
e
r
o
f
clu
s
ter
s
.
I
n
o
r
d
er
to
g
iv
e
s
tr
o
n
g
an
d
o
b
jectiv
e
esti
m
ates
o
f
th
e
p
e
r
f
o
r
m
an
ce
,
K
-
f
o
l
d
cr
o
s
s
-
v
alid
atio
n
is
u
s
e
d
o
n
a
s
er
ies
o
f
d
ata
s
p
lits
.
T
h
e
s
et
o
f
f
ea
tu
r
es
(
X)
an
d
lab
els
o
f
th
e
s
am
e
(
y
)
ar
e
d
iv
i
d
ed
in
t
o
a
tr
ain
in
g
(
Xtr
ain
,
y
tr
ain
)
an
d
test
(
Xtest,
y
test
)
s
et.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
4
3
,
No
.
1
,
Ju
ly
20
2
6
:
299
-
31
3
304
E
p
s
ilo
n
-
g
r
ee
d
y
s
tr
ateg
y
is
tak
en
to
p
ick
th
e
b
est
ac
tio
n
an
d
tar
g
et
Q
-
v
alu
es
ar
e
ca
lc
u
lated
in
B
ellm
an
eq
u
atio
n
.
T
h
e
ev
alu
a
tio
n
m
etr
ic
co
n
s
is
ts
o
f
ac
c
u
r
ac
y
an
d
F1
-
s
co
r
e,
a
n
d
,
tem
p
o
r
ar
ily
,
is
im
p
lem
en
te
d
b
y
g
en
er
atin
g
th
e
f
ee
d
b
ac
k
s
ig
n
al
r
t
o
f
t
h
e
a
g
en
t
u
s
in
g
g
r
o
u
n
d
-
tr
u
th
lab
els.
T
h
e
leg
alit
y
o
f
th
e
ap
p
r
o
ac
h
is
th
e
lack
o
f
c
o
n
tr
o
l
o
v
e
r
th
e
d
e
cisi
o
n
s
m
ad
e
b
y
th
e
ag
en
t
in
h
is
o
p
er
atio
n
,
wh
er
e
lear
n
in
g
is
p
r
o
m
o
ted
o
n
ly
b
y
lab
elled
in
f
o
r
m
atio
n
.
On
ce
t
h
e
tr
ain
in
g
e
n
d
s
,
a
co
n
f
u
s
io
n
tab
le
will
b
e
c
r
ea
t
ed
t
o
d
eter
m
in
e
th
e
lev
el
o
f
ef
f
icien
cy
o
f
th
e
s
y
s
tem
in
s
ep
ar
atin
g
n
o
r
m
al
a
n
d
m
alicio
u
s
n
etwo
r
k
tr
a
f
f
ic.
T
h
e
ad
a
p
tiv
e
clu
s
ter
in
g
is
in
tu
r
n
in
teg
r
ated
s
o
th
at
it
ca
n
d
ea
l
with
s
u
ch
a
h
ar
s
h
class
im
b
alan
ce
o
f
th
e
N
-
B
aI
o
T
d
ata
s
et
as
th
e
d
y
n
am
ic
K
wo
u
ld
all
o
w
th
e
ag
en
t
t
o
d
et
ec
t
m
in
o
r
p
atter
n
s
o
f
attac
k
t
h
at
co
u
ld
b
e
o
v
er
lo
o
k
ed
b
y
g
lo
b
al
class
if
icatio
n
.
T
h
e
b
atch
n
o
r
m
aliza
tio
n
a
n
d
d
r
o
p
o
u
t
e
x
p
er
im
e
n
tal
test
s
wer
e
d
o
n
e,
b
u
t
n
eith
er
o
f
th
e
two
f
ac
ilit
ated
co
n
v
er
g
en
ce
s
p
ee
d
o
r
s
tab
il
ity
s
ig
n
if
ican
tly
.
T
h
ey
wer
e
o
m
itted
to
m
ak
e
th
e
ar
ch
i
tectu
r
e
s
im
p
le
an
d
co
m
p
u
tatio
n
ally
ef
f
icien
t to
t
h
e
ed
g
e
d
e
v
ices o
f
th
e
I
o
T
with
o
u
t sacr
if
icin
g
th
e
d
etec
tio
n
p
er
f
o
r
m
a
n
ce
.
T
h
e
p
r
o
p
o
s
ed
alg
o
r
ith
m
was
co
d
ed
in
Py
th
o
n
with
th
e
h
el
p
o
f
th
e
T
en
s
o
r
Flo
w
f
r
am
ewo
r
k
,
a
n
d
t
h
e
k
ey
co
m
p
o
n
e
n
ts
o
f
th
e
s
im
p
le
r
ein
f
o
r
ce
m
en
t
lear
n
in
g
o
f
ex
p
er
ien
ce
r
ep
lay
,
r
ewa
r
d
s
elec
ti
o
n
,
ac
tio
n
s
elec
tio
n
p
o
licy
,
an
d
s
u
cc
ess
iv
e
u
p
d
ate
s
o
f
Q
-
n
etwo
r
k
s
wer
e
u
tili
ze
d
.
T
h
e
in
p
u
t
f
ile
was
N
-
B
aI
o
T
d
ata,
an
d
th
e
K
-
m
ea
n
s
p
r
o
v
is
io
n
in
g
m
o
d
el
th
at
was
u
tili
ze
d
wa
s
em
p
lo
y
ed
to
d
r
aw
a
r
elatio
n
b
etwe
en
th
e
ac
tio
n
s
o
f
th
e
ag
en
t.
T
r
ain
in
g
was
ca
r
r
ied
o
u
t
o
n
1
0
0
ep
is
o
d
es,
an
d
p
er
f
o
r
m
an
ce
was
e
v
alu
ated
b
y
u
s
in
g
th
e
1
0
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
in
o
r
d
er
t
o
ad
d
s
tr
en
g
th
a
n
d
s
tab
ilit
y
to
it.
Fig
u
r
e
4
also
d
em
o
n
s
tr
ates
th
e
o
v
er
all
wo
r
k
f
l
o
w
a
n
d
lo
o
p
o
f
in
ter
ac
tio
n
b
etwe
en
t
h
e
ag
en
t
a
n
d
en
v
ir
o
n
m
en
t,
in
clu
d
in
g
th
e
ch
o
ice
o
f
ac
tio
n
,
o
b
s
er
v
atio
n
o
f
th
e
s
itu
atio
n
,
an
d
Q
-
v
alu
e
u
p
d
ate
th
r
o
u
g
h
th
e
n
eu
r
al
n
etwo
r
k
.
T
h
e
m
o
d
el
ca
n
co
m
m
u
n
icate
with
th
e
I
n
ter
n
et
-
of
-
t
h
in
g
s
e
n
v
ir
o
n
m
en
t
b
y
m
o
n
ito
r
in
g
th
e
cu
r
r
en
t
s
itu
atio
n
o
f
th
e
n
etwo
r
k
tr
af
f
ic
(
S)
a
n
d
m
ak
in
g
a
n
ac
tio
n
ch
o
ice
(
A)
,
wh
ich
,
ac
c
o
r
d
i
n
g
to
an
e
-
g
r
ee
d
y
ex
p
lo
r
atio
n
p
o
licy
,
e
v
alu
ates
a
clu
s
ter
s
ize
o
f
KK.
T
o
r
ew
ar
d
th
e
ag
e
n
t,
a
r
ew
ar
d
(
R
)
is
o
b
tain
e
d
wh
e
n
th
e
ch
o
s
en
ac
tio
n
is
im
p
lem
en
ted
b
ased
o
n
th
e
m
etr
ics o
f
class
if
icatio
n
p
er
f
o
r
m
an
ce
,
e.
g
.
,
ac
c
u
r
ac
y
an
d
F1
-
s
co
r
e.
E
ac
h
ex
p
er
ien
ce
is
ad
d
e
d
to
a
r
ep
lay
b
u
f
f
er
,
wh
ich
is
s
u
b
s
e
q
u
en
tly
u
tili
ze
d
to
u
p
d
ate
t
h
e
weig
h
ts
o
f
th
e
Q
-
n
etwo
r
k
an
d
en
h
an
ce
th
e
lea
r
n
in
g
p
r
o
ce
s
s
b
y
m
ak
in
g
it
m
o
r
e
s
tab
le
an
d
co
n
v
e
r
g
en
t.
T
h
e
way
th
e
ag
en
t
p
r
o
v
id
es
f
ee
d
b
ac
k
o
n
th
e
r
e
war
d
,
r
e
-
u
p
d
ates
Q
-
v
alu
es,
b
eh
av
es
in
th
e
en
v
i
r
o
n
m
e
n
t,
a
n
d
s
elec
ts
th
e
n
ex
t
ac
tio
n
is
f
ir
s
t
d
escr
ib
ed
i
n
d
e
tail
in
Fig
u
r
e
4
,
an
d
illu
s
tr
at
e
s
th
e
DQL
tr
ain
i
n
g
lo
o
p
.
T
h
is
tr
ain
in
g
p
r
o
ce
s
s
in
v
o
lv
es
ex
p
er
ien
ce
r
e
p
lay
a
n
d
tr
ies
to
b
alan
ce
b
etwe
en
ex
p
lo
r
atio
n
an
d
ex
p
lo
itatio
n
,
s
o
th
at
it
ca
n
allo
w
o
p
tim
al
lear
n
in
g
o
f
t
h
e
attac
k
p
atter
n
s
an
d
r
esp
o
n
s
iv
en
ess
to
n
o
v
el
attac
k
p
atter
n
s
.
Fig
u
r
e
4
.
I
te
r
ativ
e
tr
ain
in
g
lo
o
p
u
s
in
g
co
m
p
o
s
ite
r
ewa
r
d
f
ee
d
b
ac
k
2
.
5
.
Det
ec
t
io
n a
nd
f
lo
w
o
f
I
o
T
tr
a
f
f
ic
T
h
e
p
r
o
p
o
s
ed
DDo
S
attac
k
-
d
etec
tin
g
s
y
s
tem
is
d
ep
lo
y
ed
a
s
a
DQL
-
b
ased
tr
a
f
f
ic
class
if
ier
th
at
h
as
th
e
p
o
ten
tial to
s
o
r
t t
h
e
n
o
r
m
a
l a
n
d
m
alicio
u
s
tr
af
f
ic.
T
h
e
o
v
er
all
p
r
o
ce
s
s
o
f
th
e
m
o
d
el
is
d
ep
icted
in
Fig
u
r
e
5
.
Fig
u
r
e
5
.
DQL
-
b
ased
m
o
d
el
f
o
r
DDo
S
de
tectio
n
in
I
o
T
t
r
af
f
ic
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
Dee
p
Q
lea
r
n
in
g
a
l
g
o
r
ith
m
fo
r
d
etec
tin
g
DDo
S
a
tta
ck
s
o
n
I
o
T d
ev
ices
(
La
n
a
K
a
mla
A
h
m
ed
)
305
2
.
6
.
E
x
perim
ent
a
l
e
qu
ipm
e
nt
Py
th
o
n
3
.
1
2
.
3
was
u
s
ed
as
th
e
im
p
lem
en
tatio
n
lan
g
u
ag
e
f
o
r
th
e
p
r
o
p
o
s
ed
m
o
d
el,
en
a
b
lin
g
s
cien
tific
co
m
p
u
tin
g
an
d
m
ac
h
i
n
e
-
lear
n
in
g
-
b
ased
r
esear
ch
in
a
m
o
d
e
r
n
en
v
ir
o
n
m
en
t.
T
e
n
s
o
r
Flo
w
2
.
1
7
.
0
was
u
s
ed
to
d
ev
elo
p
an
d
tr
ain
th
e
DQL
m
o
d
el,
as
it
is
m
atu
r
e
an
d
co
m
p
atib
le
with
T
en
s
o
r
Flo
w
L
ite,
e
n
ab
lin
g
d
ep
lo
y
m
en
t
o
n
r
eso
u
r
ce
-
lim
ited
I
o
T
ed
g
e
d
ev
ices.
Scik
it
-
lear
n
was
u
s
ed
to
p
r
ep
r
o
ce
s
s
an
d
clu
s
ter
d
ata
with
K
-
m
ea
n
s
,
wh
ile
Nu
m
Py
an
d
Pan
d
as
ass
is
ted
with
d
ata
m
an
ip
u
latio
n
,
an
d
Ma
tp
l
o
tlib
was
u
s
ed
f
o
r
p
e
r
f
o
r
m
an
ce
v
is
u
aliza
tio
n
.
All
ca
lcu
latio
n
s
wer
e
p
er
f
o
r
m
e
d
o
n
a
m
u
lti
-
c
o
r
e
I
n
tel
C
o
r
e
i7
-
1
0
7
5
0
H
C
PU
at
2
.
6
0
GHz
with
1
6
GB
o
f
R
AM
an
d
a
5
1
2
G
B
SS
D,
s
im
u
latin
g
a
r
ea
lis
tic
h
ig
h
-
e
n
d
I
o
T
ed
g
e
g
atew
ay
.
T
h
e
ex
p
e
r
im
en
tal
p
r
o
ce
s
s
in
v
o
lv
e
d
p
r
o
ce
s
s
in
g
9
1
d
is
tin
ct
d
ata
f
iles
f
r
o
m
th
e
N
-
B
aI
o
T
d
ataset.
T
h
e
m
ea
n
c
r
o
s
s
-
v
alid
atio
n
tim
e
o
f
6
5
.
6
s
ec
o
n
d
s
,
o
r
6
5
.
6
s
ec
o
n
d
s
p
er
f
o
ld
o
f
cr
o
s
s
-
v
alid
atio
n
,
im
p
lied
a
to
tal
tr
ain
in
g
a
n
d
cr
o
s
s
-
v
alid
atio
n
tim
e
o
f
ab
o
u
t
3
.
9
h
o
u
r
s
in
t
h
e
f
u
ll
c
o
u
r
s
e
o
f
th
e
1
0
-
f
o
ld
cr
o
s
s
v
alid
atio
n
.
T
h
e
h
i
g
h
est
m
e
m
o
r
y
co
n
s
u
m
p
tio
n
was m
ain
tain
ed
at
5
0
0
MB,
an
d
th
e
av
er
ag
e
i
n
f
er
en
ce
tim
e
was 0
.
0
5
m
s
p
er
in
s
tan
ce
.
T
h
e
s
e
v
alu
es v
er
if
y
th
at
th
e
m
o
d
el
is
co
m
p
u
tatio
n
ally
ef
f
icien
t a
n
d
ca
n
b
e
u
s
ed
in
lo
w
-
r
eso
u
r
ce
an
d
lo
w
-
m
em
o
r
y
I
o
T
ap
p
licatio
n
s
.
2
.
7
.
P
er
f
o
r
m
a
nce
m
e
t
rics
Fo
u
r
p
o
p
u
lar
m
etr
ics,
n
am
el
y
,
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
an
d
F1
-
s
co
r
e,
wer
e
u
s
ed
to
ass
es
s
th
e
class
if
icatio
n
p
er
f
o
r
m
an
ce
o
f
th
e
p
r
o
p
o
s
ed
DQL
m
o
d
el
[
3
3
]
.
Su
ch
m
ea
s
u
r
es
ca
n
b
e
u
s
ed
to
d
eter
m
in
e
th
e
ef
f
ec
tiv
en
ess
o
f
th
e
m
o
d
el
i
n
d
ec
id
in
g
th
e
p
r
esen
ce
o
f
m
alig
n
an
t
an
d
leg
itima
te
n
et
wo
r
k
tr
af
f
ic
in
th
e
d
etec
tio
n
o
f
I
o
T
-
b
ased
DDo
S.
Acc
u
r
ac
y
is
a
m
ea
s
u
r
e
o
f
th
e
f
r
ac
tio
n
o
f
tr
u
e
s
am
p
les in
a
s
et
o
f
s
am
p
les:
A
c
c
ura
c
y
=
+
+
+
+
(
1
)
T
h
e
ac
cu
r
ac
y
is
a
m
ea
s
u
r
e
o
f
th
e
n
u
m
b
er
o
f
p
o
s
itiv
es
th
a
t
ar
e
p
r
e
d
icted
co
r
r
ec
tly
o
u
t
o
f
all
th
e
p
r
ed
icted
p
o
s
itiv
es:
Pr
e
c
ision
=
+
(
2
)
R
ec
all:
R
ec
all
i
s
a
r
atio
th
at
is
u
s
ed
to
m
ea
s
u
r
e
h
o
w
m
an
y
p
o
s
itiv
e
ca
s
es
ar
e
co
r
r
e
ctly
id
e
n
tifie
d
o
u
t
o
f
all
th
e
tr
u
e
p
o
s
itiv
e
ca
s
es:
R
e
c
a
l
l
=
+
(
3
)
F1
-
s
co
r
e
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:
1
=
2
×
Pre
c
isi
o
n
×
Re
c
a
l
l
Pre
c
isi
o
n
+
Re
c
a
l
l
(
4
)
All
o
f
th
ese
m
etr
ics
r
ep
r
esen
t
a
s
tr
o
n
g
s
y
s
tem
th
at
ca
n
b
e
u
s
ed
to
ev
alu
ate
th
e
d
etec
tio
n
p
o
ten
tial
o
f
th
e
m
o
d
el
in
p
a
r
ticu
lar
ca
s
es
o
f
im
b
alan
ce
d
d
ata,
wh
en
t
h
e
p
r
esen
ce
o
f
attac
k
tr
af
f
ic
s
am
p
l
es
m
ay
o
v
er
wh
elm
th
e
n
o
r
m
a
l
d
ata
tr
af
f
ic.
T
h
ese
m
etr
ics
ar
e
ap
p
lied
in
o
r
d
er
to
co
m
p
ar
e
th
e
p
r
o
p
o
s
ed
DQL
-
K
-
m
ea
n
s
m
o
d
el
with
m
u
ltip
le
b
aselin
es,
s
u
ch
as
R
an
d
o
m
Fo
r
est,
C
NN,
an
d
Q
-
L
ea
r
n
in
g
,
in
o
r
d
e
r
to
g
u
a
r
a
n
tee
m
eth
o
d
o
lo
g
ical
r
ig
o
r
.
Als
o
,
an
ab
latio
n
s
tu
d
y
is
ca
r
r
ied
o
u
t
with
th
e
u
s
e
o
f
t
h
e
s
am
e
m
etr
ics
to
ch
ec
k
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
m
o
d
el
wh
e
n
th
e
ad
ap
tiv
e
clu
s
t
er
in
g
a
n
d
t
h
e
elem
en
ts
o
f
c
o
m
p
o
s
ite
r
ewa
r
d
ar
e
a
b
s
en
t.
All
r
esu
lts
ar
e
g
iv
en
as
th
e
m
ea
n
o
f
th
e
1
0
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
f
o
ld
s
in
o
r
d
er
to
co
n
s
id
er
s
tatis
tical
v
ar
i
ab
ilit
y
a
n
d
o
b
tain
s
ig
n
if
ican
ce
,
alo
n
g
with
th
e
s
tan
d
ar
d
d
ev
iatio
n
(
±
σ
)
.
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
I
n
th
is
s
ec
tio
n
,
th
e
r
esu
lts
o
f
th
e
ex
p
e
r
im
en
t
o
f
th
e
p
r
o
p
o
s
ed
DQL
-
b
ased
f
r
am
ewo
r
k
a
r
e
p
r
o
v
id
ed
.
T
h
is
ev
alu
atio
n
h
as
b
ee
n
p
er
f
o
r
m
ed
b
ased
o
n
th
e
N
-
B
aI
o
T
d
ata
s
et
an
d
th
e
co
m
m
o
n
p
e
r
f
o
r
m
an
ce
m
ea
s
u
r
es
o
f
th
e
N
-
B
aI
o
T
:
ac
cu
r
ac
y
,
F1
-
s
co
r
e,
tr
u
e
p
o
s
itiv
es
(
T
P),
tr
u
e
n
e
g
ativ
es
(
T
N)
,
f
alse
p
o
s
it
iv
es
(
FP
)
,
an
d
f
alse
n
eg
ativ
es
(
FN)
.
T
h
e
co
n
v
er
g
en
ce
v
is
u
aliza
tio
n
s
an
d
th
e
t
em
p
o
r
al
p
r
o
g
r
ess
io
n
o
f
th
e
m
o
d
el
d
escr
ib
e
th
e
lear
n
in
g
p
r
o
ce
s
s
o
f
th
e
m
o
d
el
in
a
s
er
ies
o
f
ep
is
o
d
es
an
d
d
is
p
lay
s
tab
ilit
y
an
d
s
tab
le
co
n
v
er
g
en
ce
b
eh
av
i
o
u
r
.
Als
o
,
m
ea
s
u
r
es
o
f
co
m
p
u
tati
o
n
al
ef
f
icien
cy
we
r
e
m
ad
e
i
n
ter
m
s
o
f
tim
e
(
in
s
ec
o
n
d
s
)
tak
en
to
tr
ain
an
ex
am
p
le,
p
er
-
s
am
p
le
in
f
er
en
c
e
laten
cy
,
an
d
m
em
o
r
y
co
n
s
u
m
p
tio
n
.
All
th
ese
f
in
d
in
g
s
ar
e
a
co
m
p
r
eh
e
n
s
iv
e
r
ev
iew
o
f
th
e
ef
f
ec
tiv
en
ess
an
d
ef
f
icien
cy
o
f
th
e
s
u
g
g
ested
DQL
m
o
d
el.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
4
3
,
No
.
1
,
Ju
ly
20
2
6
:
299
-
31
3
306
3
.
1
.
Cla
s
s
if
ica
t
io
n
perf
o
r
m
a
nce
re
s
ults
Acc
u
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
a
n
d
F1
-
s
co
r
e
ar
e
f
o
u
r
s
tan
d
a
r
d
m
ea
s
u
r
es
th
at
wer
e
em
p
lo
y
ed
to
test
th
e
p
r
o
p
o
s
ed
DQL
m
o
d
el.
T
h
e
v
alu
es
u
n
d
er
th
ese
m
etr
ics
h
av
e
b
ee
n
ca
lcu
lated
th
r
o
u
g
h
t
h
e
elem
en
ts
o
f
th
e
co
n
f
u
s
io
n
m
atr
ices
b
ec
a
u
s
e
t
h
e
elem
en
ts
h
av
e
b
ee
n
s
u
m
m
ar
ized
in
T
ab
le
3
[
3
3
]
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
h
as
d
em
o
n
s
tr
ated
a
h
ig
h
lev
el
o
f
d
etec
tio
n
an
d
an
ac
ce
p
tab
le
p
r
o
p
o
r
ti
o
n
o
f
p
r
ec
is
io
n
an
d
r
ec
all,
as
p
r
esen
ted
in
th
e
r
esu
lts
o
f
t
h
e
ex
p
er
im
en
t.
T
h
ese
v
a
lu
es
o
f
F1
-
s
co
r
e
ar
e
also
an
e
x
ce
llen
t
in
d
icatio
n
t
h
at
th
e
m
o
d
el
ca
n
d
is
tin
g
u
is
h
b
etwe
en
th
e
m
alicio
u
s
tr
af
f
ic
an
d
th
e
r
ea
l
I
o
T
n
etwo
r
k
tr
af
f
ic.
Fig
u
r
e
6
s
h
o
w
s
th
e
d
is
tr
ib
u
tio
n
o
f
r
ea
lized
an
d
esti
m
ated
class
es
o
f
a
n
a
v
er
ag
e
f
o
ld
o
f
th
e
c
r
o
s
s
-
v
alid
atio
n
p
r
o
ce
d
u
r
e
o
f
1
0
f
o
ld
s
.
Fig
u
r
e
6
in
d
icate
s
th
at
f
alse
p
o
s
itiv
es
wer
e
lo
w
(
2
1
,
6
3
5
)
in
th
e
m
o
d
el.
T
h
is
is
ess
en
tial
to
I
o
T
en
v
ir
o
n
m
en
t
s
in
ce
h
ig
h
f
alse
p
o
s
itiv
e
will
r
esu
lt
in
b
lo
ck
in
g
o
f
leg
itima
te
tr
af
f
ic,
lik
e
s
m
ar
t
s
en
s
o
r
s
u
p
d
ates
o
r
u
s
er
co
m
m
an
d
s
,
to
b
r
in
g
d
o
wn
th
e
s
er
v
ice.
On
th
e
o
th
er
h
an
d
,
th
e
Fals
e
Neg
ati
v
e
(
9
,
8
9
6
)
is
lo
w,
wh
ich
m
ea
n
s
th
at
th
e
m
ajo
r
ity
o
f
m
alicio
u
s
DDo
S
tr
af
f
ic
will
b
e
ap
p
lied
p
r
i
o
r
to
d
am
ag
i
n
g
th
e
eq
u
ip
m
en
t.
E
x
am
in
in
g
th
ese
o
v
er
lo
o
k
ed
in
s
tan
ce
s
o
f
attac
k
s
h
as
s
h
o
wn
th
at
th
e
m
o
d
el
s
o
m
etim
es
h
as
d
if
f
icu
lties
s
tealth
y
U
DP
s
ca
n
n
in
g
,
wh
ic
h
esch
ews
th
e
h
ig
h
-
f
r
eq
u
en
cy
o
f
th
e
h
ea
r
tb
ea
t
s
ig
n
als
o
f
m
al
icio
u
s
I
o
T
d
ev
ices.
T
h
e
ad
a
p
tiv
e
q
u
ality
o
f
th
e
DQL
ag
en
t
h
o
wev
er
ass
is
ts
i
n
r
ed
u
cin
g
th
is
b
y
ch
an
g
i
n
g
th
e
clu
s
ter
r
ep
r
esen
tatio
n
as
th
e
b
eh
av
io
r
o
f
th
e
attac
k
ad
ap
ts
.
T
ab
le
3
.
B
in
ar
y
class
if
icatio
n
p
er
f
o
r
m
an
ce
o
f
th
e
c
o
n
f
u
s
io
n
m
atr
ix
P
r
e
d
i
c
t
e
d
p
o
si
t
i
v
e
P
r
e
d
i
c
t
e
d
n
e
g
a
t
i
v
e
A
c
t
u
a
l
p
o
si
t
i
v
e
4
,
5
0
0
,
0
0
0
9
,
8
9
6
A
c
t
u
a
l
n
e
g
a
t
i
v
e
2
1
,
6
3
5
2
,
0
0
0
,
0
0
0
Fig
u
r
e
6
.
C
o
n
f
u
s
io
n
m
atr
i
x
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
,
wh
ich
h
a
s
h
ig
h
v
alu
es o
f
class
if
icatio
n
an
d
lo
w
v
al
u
es in
er
r
o
r
s
3
.
2
.
Ana
ly
s
is
o
f
t
r
a
ini
ng
ev
o
lutio
n a
nd
co
nv
er
g
ence
On
th
e
s
tab
ilit
y
o
f
th
e
p
r
o
p
o
s
ed
DQL
m
o
d
el,
we
p
r
esen
t
m
ea
s
u
r
es
o
f
p
er
f
o
r
m
an
ce
th
r
o
u
g
h
th
e
co
u
r
s
e
o
f
all
ep
is
o
d
es
an
d
f
o
l
d
s
o
f
a
ten
-
f
o
l
d
cr
o
s
s
-
v
alid
ati
o
n
s
tr
ateg
y
.
Als
o
,
in
tr
ain
in
g
an
d
test
in
g
,
ac
cu
r
ac
y
an
d
F1
-
s
co
r
e
wer
e
em
p
lo
y
e
d
to
test
th
e
d
y
n
am
ics
o
f
lear
n
in
g
in
th
e
c
o
u
r
s
e
o
f
tim
e.
Fig
u
r
e
7
s
h
o
ws
th
e
tr
ain
in
g
d
ev
elo
p
m
en
t,
i
n
wh
i
ch
th
e
ac
c
u
r
ac
y
an
d
F1
-
s
co
r
e
lev
el
o
f
f
ea
r
ly
d
u
r
in
g
th
e
d
e
v
elo
p
m
en
t.
Su
ch
a
q
u
ick
co
n
v
er
g
en
ce
is
ex
p
lain
e
d
b
y
th
e
co
m
p
o
s
ite
r
ewa
r
d
f
u
n
ctio
n
th
at
g
iv
es r
ich
f
ee
d
b
ac
k
to
th
e
ag
en
t so
th
at
it
ca
n
r
ap
id
ly
d
if
f
er
en
tiate
b
et
wee
n
b
en
i
g
n
tr
af
f
ic
an
d
d
i
f
f
er
en
t
DDo
S
attac
k
v
ec
to
r
s
in
th
e
N
-
B
aI
o
T
d
ata
s
et.
T
h
ese
m
ea
s
u
r
em
en
ts
ten
d
to
b
e
ex
tr
em
ely
elev
ated
in
m
aj
o
r
ity
o
f
th
e
ep
is
o
d
es
an
d
g
r
ea
tly
v
ar
y
b
ec
au
s
e
o
f
p
er
s
o
n
al
d
is
cr
ep
an
cy
in
th
e
ex
p
er
im
en
tal
f
u
n
ctio
n
in
g
o
f
th
e
DQL
alg
o
r
ith
m
.
Su
cc
ess
f
u
l
lear
n
in
g
ca
n
b
e
attr
ib
u
ted
to
th
is
tr
en
d
,
an
d
it sh
o
ws th
at
th
e
m
o
d
el
is
co
n
v
er
g
en
t in
n
at
u
r
e.
Fig
u
r
e
7
.
DQL
t
r
ain
i
n
g
f
o
r
a
cc
u
r
ac
y
an
d
F1
s
co
r
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
Dee
p
Q
lea
r
n
in
g
a
l
g
o
r
ith
m
fo
r
d
etec
tin
g
DDo
S
a
tta
ck
s
o
n
I
o
T d
ev
ices
(
La
n
a
K
a
mla
A
h
m
ed
)
307
Fo
llo
win
g
th
e
tr
ain
in
g
ev
alu
at
io
n
,
p
er
f
o
r
m
a
n
ce
m
ea
s
u
r
es
o
f
test
in
g
ar
e
illu
s
tr
ated
in
Fig
u
r
e
8
,
wh
er
e
ac
cu
r
ac
y
a
n
d
F1
-
s
co
r
e
ar
e
p
lo
tted
with
th
e
v
ar
io
u
s
f
o
ld
s
o
f
c
r
o
s
s
-
v
alid
atio
n
o
f
th
e
test
in
g
.
Fig
u
r
e
8
r
ep
r
esen
ts
th
e
test
in
g
p
er
f
o
r
m
a
n
ce
o
f
th
e
1
0
-
f
o
ld
c
r
o
s
s
-
v
alid
atio
n
.
T
h
e
f
ac
t
th
at
th
e
v
ar
ian
ce
am
o
n
g
th
e
f
o
ld
s
is
m
in
im
al
(
it
is
o
b
s
er
v
e
d
in
th
e
tig
h
t
clu
s
ter
in
g
o
f
th
e
d
a
ta
p
o
in
ts
)
p
r
o
v
es
th
e
f
ac
t
t
h
at
th
e
m
o
d
el
h
as g
o
o
d
g
e
n
er
aliza
tio
n
p
o
wer
.
T
h
is
will
m
ak
e
th
e
s
y
s
tem
ef
f
ec
tiv
e
wh
en
n
ew,
u
n
k
n
o
wn
I
o
T
d
e
v
ices
o
r
d
if
f
e
r
en
t
co
n
d
itio
n
s
o
f
n
etwo
r
k
tr
af
f
ic
ar
e
in
v
o
lv
ed
,
wh
ich
wer
e
n
o
t e
x
p
er
ie
n
ce
d
d
u
r
in
g
th
e
tr
ain
in
g
f
o
ld
.
T
h
e
an
aly
s
is
o
f
th
e
p
r
o
g
r
ess
io
n
o
f
th
e
m
ax
im
u
m
Q
-
v
alu
e
s
o
f
th
e
ag
en
t
o
v
er
ep
is
o
d
es
was
al
s
o
m
o
n
ito
r
ed
i
n
th
e
ex
p
er
im
en
t
to
ex
am
in
e
co
n
v
er
g
en
ce
.
C
o
n
v
er
g
e
n
ce
o
f
th
e
lea
r
n
in
g
p
r
o
ce
s
s
is
as
s
u
m
ed
to
h
av
e
b
ee
n
s
u
cc
ess
f
u
l
in
th
e
ev
en
t
o
f
a
co
n
s
tan
t
o
f
s
tab
le
tr
en
d
in
t
h
e
Q
-
v
al
u
es;
th
e
tr
en
d
in
th
is
ca
s
e
is
p
r
esen
ted
in
Fig
u
r
e
9
.
Fig
u
r
e
8
.
Acc
u
r
ac
y
a
n
d
F1
-
s
c
o
r
e
f
o
l
d
s
f
o
r
test
in
g
Fig
u
r
e
9
.
Ma
x
im
u
m
Pre
d
icted
Q
-
v
alu
e
Per
T
r
ai
n
in
g
E
p
is
o
d
e
T
im
e
d
y
n
a
m
ics
o
f
th
e
m
a
x
i
m
u
m
Q
-
v
alu
es
in
s
eq
u
en
tial
tr
ain
in
g
s
ess
io
n
s
we
r
e
s
ee
n
to
test
th
e
s
tab
ilit
y
o
f
th
e
lear
n
in
g
p
r
o
ce
s
s
.
F
ig
u
r
e
9
s
h
o
ws
th
e
m
ax
im
u
m
p
r
ed
icted
Q
-
v
alu
es
ar
e
v
a
r
y
in
g
in
tr
a
-
e
p
is
o
d
e
as
o
p
p
o
s
ed
to
a
s
tr
aig
h
t
lin
e.
T
h
is
is
a
g
o
o
d
s
ig
n
th
at
th
e
ep
s
ilo
n
-
d
ec
ay
m
ec
h
a
n
is
m
is
b
eh
av
in
g
p
r
o
p
er
ly
,
th
e
ag
en
t
k
ee
p
s
s
ea
r
ch
i
n
g
a
h
i
g
h
-
d
im
en
s
io
n
al
s
p
ac
e
o
f
I
o
T
tr
af
f
ic
with
o
u
t
f
allin
g
i
n
to
th
e
lo
ca
l
o
p
tim
a
o
f
t
h
e
s
tr
ateg
y
,
s
o
th
at
th
e
r
esu
ltin
g
p
o
licy
will b
e
r
esis
tan
t
to
s
tealth
y
,
a
d
ap
tab
le
attac
k
e
x
ec
u
tio
n
s
.
No
n
eth
eless
,
th
e
ac
cu
r
ac
y
an
d
F1
-
s
co
r
e
cu
r
v
es in
Fig
u
r
e
9
d
o
n
o
t d
if
f
er
b
etw
ee
n
th
e
cr
o
s
s
-
v
alid
atio
n
f
o
ld
s
,
wh
ich
in
d
icate
s
th
e
p
r
esen
ce
o
f
g
o
o
d
r
esu
lts
in
ter
m
s
o
f
lear
n
in
g
.
T
h
e
cr
o
s
s
-
v
alid
atio
n
f
o
ld
r
e
q
u
ir
ed
a
p
p
r
o
x
im
ately
6
5
.
6
s
ec
o
n
d
s
o
f
wall
-
clo
ck
tim
e;
th
e
m
ea
n
in
f
er
en
ce
tim
e
was
0
.
0
5
m
s
p
e
r
in
s
tan
ce
,
a
n
d
t
h
e
m
o
d
el
o
cc
u
p
an
cy
was
ar
o
u
n
d
5
0
0
MB,
in
d
icatin
g
th
e
co
m
p
u
tatio
n
al
ef
f
icien
cy
o
f
th
e
m
o
d
el
an
d
m
ak
in
g
it
a
u
s
ef
u
l
in
s
tan
ce
to
b
e
u
s
ed
in
lo
w
-
m
em
o
r
y
I
o
T
ap
p
licatio
n
s
.
3
.
3
.
Q
-
lea
rning
re
s
ults
T
h
e
Q
-
lear
n
in
g
m
o
d
el
p
r
o
p
o
s
ed
h
as
r
ea
ch
ed
th
e
ac
cu
r
ac
y
o
f
th
e
class
if
icatio
n
o
f
9
2
.
3
1
,
p
r
ec
is
io
n
o
f
9
2
.
3
1
,
r
ec
all
o
f
9
2
.
1
3
,
an
d
F1
-
s
co
r
e
o
f
9
2
.
4
0
b
y
ad
j
u
s
tin
g
t
h
e
lear
n
in
g
r
ate
(
0
.
0
5
)
an
d
d
is
co
u
n
t
f
ac
to
r
(
0
.
9
5
)
o
v
er
th
e
n
u
m
b
er
o
f
iter
atio
n
s
.
T
o
f
in
d
th
e
b
e
st
-
k
n
o
wn
p
ar
a
m
eter
s
,
th
e
v
alu
es
o
f
th
e
h
y
p
e
r
p
ar
am
eter
s
α
an
d
γ
wer
e
o
b
tain
ed
th
r
o
u
g
h
t
h
e
i
n
tu
itiv
e
(
n
o
n
-
co
m
p
u
tatio
n
al
)
s
ea
r
ch
o
f
a
s
elec
ted
h
y
p
er
p
a
r
am
eter
s
p
ac
e,
b
y
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
4
3
,
No
.
1
,
Ju
ly
20
2
6
:
299
-
31
3
308
ass
es
s
in
g
th
e
v
alid
atio
n
p
er
f
o
r
m
an
ce
.
Su
b
s
tan
tial
en
h
a
n
ce
m
en
t
o
f
th
e
u
n
tu
n
ed
b
aselin
e
was
o
b
tain
ed
b
y
th
is
tu
n
in
g
p
r
o
ce
s
s
.
I
n
ad
d
itio
n
,
1
0
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
(
K
=
1
0
)
u
s
in
g
th
e
o
p
tim
ized
h
y
p
er
p
ar
am
eter
s
r
esu
lted
in
an
ac
c
u
r
ac
y
o
f
9
5
.
2
4
,
p
r
ec
i
s
io
n
o
f
9
5
.
0
2
,
r
ec
all
o
f
9
5
.
2
4
,
an
d
an
F1
-
s
co
r
e
o
f
9
5
.
1
3
.
T
a
b
le
4
also
g
iv
es
a
m
o
r
e
s
p
ec
if
ic
co
m
p
a
r
is
o
n
o
f
ev
alu
atio
n
m
ea
s
u
r
es
o
f
th
e
p
r
o
p
o
s
ed
Q
-
lear
n
in
g
m
o
d
el
as
is
an
d
af
ter
th
e
ap
p
licatio
n
o
f
th
e
K
-
f
o
ld
an
d
K
-
m
ea
n
s
.
T
ab
le
4
.
Q
-
lear
n
in
g
m
o
d
els r
e
s
u
lt
P
e
r
f
o
r
ma
n
c
e
m
e
t
r
i
c
s
Q
-
l
e
a
r
n
i
n
g
Q
-
l
e
a
r
n
i
n
g
(
K
-
f
o
l
d
a
n
d
K
-
m
e
a
n
)
A
c
c
u
r
a
c
y
9
2
.
3
1
%
9
5
.
2
4
%
P
r
e
c
i
s
i
o
n
9
2
.
3
1
%
9
5
.
0
2
%
R
e
c
a
l
l
9
2
.
1
3
%
9
5
.
2
4
%
F1
-
S
c
o
r
e
9
2
.
4
0
%
9
5
.
1
3
%
I
t
is
s
h
o
wn
in
th
e
tab
le
ab
o
v
e
th
at
th
e
p
er
f
o
r
m
an
ce
m
ea
s
u
r
es
o
f
th
e
Q
-
lear
n
i
n
g
m
o
d
el
an
d
th
e
m
o
d
i
f
ied
m
o
d
el
a
r
e
co
m
p
ar
e
d
,
in
clu
d
in
g
K
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
an
d
K
-
m
ea
n
s
clu
s
ter
in
g
.
T
h
e
cr
o
s
s
-
v
alid
atio
n
p
lu
s
th
e
K
-
m
ea
n
s
is
f
air
ly
s
u
cc
es
s
f
u
l
i
n
en
h
an
cin
g
all
ass
e
s
s
m
en
t
m
ea
s
u
r
es,
esp
ec
ially
th
e
r
ec
all
an
d
ac
cu
r
ac
y
.
T
h
e
r
esu
lts
o
f
th
e
co
m
p
ar
is
o
n
s
ar
e
r
ep
r
esen
ted
in
Fig
u
r
e
1
0
;
th
er
ef
o
r
e,
th
e
f
o
c
u
s
is
p
u
t o
n
t
h
e
b
etter
s
en
s
itiv
ity
o
f
th
e
m
o
d
if
ied
m
o
d
el
t
o
id
en
tif
y
th
e
in
d
iv
id
u
al.
Fig
u
r
e
1
0
d
em
o
n
s
tr
ates
th
e
in
f
lu
en
ce
o
f
h
y
p
er
p
ar
am
eter
o
p
tim
izatio
n
o
n
th
e
p
er
f
o
r
m
an
ce
o
f
a
Q
-
lear
n
in
g
m
o
d
el
to
id
en
tif
y
DDo
S
attac
k
s
s
ta
tin
g
th
at
it
g
en
er
ally
in
cr
ea
s
es
th
e
ab
ili
ty
to
d
etec
t
th
ese
attac
k
s
.
So
m
e
ex
tr
a
p
er
f
o
r
m
a
n
ce
wo
u
ld
b
e
o
b
tain
ed
with
K
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
a
n
d
K
-
Me
an
s
clu
s
ter
in
g
th
at
wo
u
ld
r
ed
u
ce
o
v
er
f
itti
n
g
a
n
d
p
r
o
v
id
e
g
r
ea
ter
r
eg
u
la
r
ity
ac
r
o
s
s
d
iv
is
io
n
s
o
f
d
ata.
Nev
e
r
th
eless
,
th
eir
s
p
ec
if
ic
d
ef
icien
cies
r
em
ain
th
e
lim
its
in
th
e
p
r
ec
is
io
n
an
d
th
eir
a
b
ilit
y
to
d
etec
t
attac
k
s
,
w
h
ich
lea
d
s
to
th
e
n
ec
ess
ity
to
in
tr
o
d
u
ce
m
o
r
e
s
o
p
h
is
ticated
ap
p
r
o
ac
h
es
th
at
will
b
e
ab
le
to
wo
r
k
o
n
th
e
n
ew
d
y
n
a
m
ica
lly
ev
o
lv
in
g
attac
k
p
atter
n
s
o
n
I
o
T
n
etwo
r
k
s
.
Fig
u
r
e
1
0
.
Q
-
lear
n
in
g
p
er
f
o
r
m
an
ce
an
d
i
n
teg
r
atin
g
(K
-
Fo
ld
a
n
d
K
-
Me
an
)
3
.
4
.
DQ
L
r
esu
lt
s
A
lear
n
in
g
r
ate
(
0
.
0
0
1
)
was
u
s
ed
s
o
th
at
th
e
weig
h
t
ch
an
g
ed
with
o
u
t
ac
ce
ler
atio
n
;
a
h
ig
h
er
v
alu
e
led
to
a
g
lo
b
b
in
g
,
an
d
a
lo
wer
v
alu
e
led
to
n
o
n
-
lea
r
n
in
g
.
W
e
h
av
e
f
ix
ed
th
e
d
is
co
u
n
t
f
ac
t
o
r
(
0
.
9
5
)
to
en
a
b
le
th
e
ag
en
t
to
ca
p
tu
r
e
f
u
tu
r
e
n
etwo
r
k
s
tates,
wh
ich
is
ap
p
r
o
p
r
iate
to
ca
p
tu
r
e
th
e
tem
p
o
r
al
d
ep
en
d
en
cies
o
f
m
u
lti
-
s
tag
e
DDo
S
attac
k
s
.
T
h
e
ep
s
ilo
n
-
d
ec
ay
s
ch
ed
u
le
b
eg
i
n
s
with
an
ep
s
ilo
n
o
f
1
.
0
an
d
d
ec
r
ea
s
es
to
0
.
0
1
in
th
e
f
ir
s
t
1
0
0
ep
is
o
d
es,
g
iv
in
g
a
p
r
elim
in
ar
y
b
r
o
a
d
s
ea
r
ch
o
f
th
e
h
ig
h
-
d
im
e
n
s
io
n
al
N
-
B
aI
o
T
p
r
o
p
e
r
ties
af
ter
wh
ich
th
e
ag
e
n
t
is
co
n
s
id
e
r
ed
to
s
witch
to
an
eq
u
ilib
r
iu
m
ex
p
l
o
itatio
n
p
o
licy
.
T
h
e
r
esu
lt
o
f
t
h
ese
o
p
tim
izatio
n
s
was
a
b
aselin
e
p
er
f
o
r
m
a
n
ce
f
o
r
th
e
m
o
d
e
l
o
f
9
5
.
8
3
%
ac
c
u
r
ac
y
,
9
5
.
8
3
%
r
ec
all,
9
3
.
7
5
%
p
r
ec
is
io
n
,
an
d
an
F1
-
s
co
r
e
o
f
9
4
.
4
4
%.
L
ater
o
p
tim
izatio
n
s
in
clu
d
ed
a
d
d
in
g
K
-
m
ea
n
s
cl
u
s
ter
in
g
to
th
e
DQL
tr
ain
in
g
p
r
o
ce
s
s
an
d
u
s
in
g
K
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
.
K
-
m
ea
n
s
clu
s
ter
in
g
was
u
s
ed
to
r
ed
u
ce
n
o
is
e
an
d
m
ak
e
class
es m
o
r
e
s
ep
ar
ab
le,
th
er
eb
y
m
ak
in
g
th
e
d
ata
m
o
r
e
a
m
en
ab
le
to
DQL
tr
ain
in
g
.
T
h
e
c
r
o
s
s
-
v
al
i
d
a
ti
o
n
p
r
o
c
ess
was
p
e
r
f
o
r
m
e
d
u
s
i
n
g
K
d
if
f
e
r
e
n
t
d
ata
p
a
r
tit
io
n
s
,
a
n
d
th
e
m
o
d
el
w
as
v
al
id
at
ed
,
p
r
o
v
i
d
i
n
g
co
n
s
is
te
n
cy
a
n
d
e
n
a
b
l
in
g
g
e
n
e
r
a
liz
ati
o
n
t
o
n
ew
d
at
a.
T
h
e
m
o
s
t
p
o
s
iti
v
e
r
es
u
lts
w
er
e
ac
h
ie
v
e
d
wi
th
t
h
e
i
n
t
eg
r
a
te
d
m
o
d
el,
w
h
i
ch
c
o
m
b
in
e
d
o
p
ti
m
iz
ed
h
y
p
er
p
a
r
am
ete
r
s
,
K
-
m
ea
n
s
cl
u
s
te
r
i
n
g
,
a
n
d
K
-
f
o
ld
cr
o
s
s
-
v
ali
d
ati
o
n
.
T
h
is
s
etu
p
ac
h
ie
v
ed
9
8
.
9
5
%
±
0
.
1
7
%
a
cc
u
r
a
cy
,
9
8
.
7
2
%
±
0
.
1
7
%
p
r
ec
is
i
o
n
,
9
8
.
9
5
%
±
0
.
1
7
%
r
ec
all
,
a
n
d
a
n
F1
-
s
c
o
r
e
o
f
9
8
.
7
3
%
±
0
.
1
7
%
,
w
h
ic
h
w
a
s
b
y
f
a
r
b
et
te
r
t
h
a
n
t
h
e
b
as
eli
n
e
an
d
t
h
e
im
p
r
o
v
ed
DQL
m
o
d
e
ls
.
A
t
-
t
est
c
o
m
p
ar
i
n
g
th
e
F
1
Sc
o
r
es
f
r
o
m
1
0
-
f
o
l
d
c
r
o
s
s
-
v
a
li
d
at
io
n
p
r
o
v
i
d
e
d
ev
id
e
n
c
e
o
f
t
h
e
s
tatis
t
ic
al
s
i
g
n
i
f
ic
a
n
ce
o
f
th
is
i
m
p
r
o
v
e
m
e
n
t
.
T
h
e
p
-
v
al
u
e
o
b
ta
in
e
d
f
r
o
m
th
e
t
est w
as
5
.
0
8
×
1
0
^
-
5
,
i
n
d
ica
ti
n
g
t
h
at
Evaluation Warning : The document was created with Spire.PDF for Python.