I
n
t
e
r
n
at
io
n
al
Jou
r
n
al
of
A
d
van
c
e
s
i
n
A
p
p
li
e
d
S
c
ie
n
c
e
s
(
I
JA
A
S
)
V
ol
. 15
, N
o
. 1
,
M
a
r
c
h
20
26
, pp
.
198
~
208
I
S
S
N
:
2252
-
8814
,
D
O
I
:
10.11591/
ij
a
a
s
.
v15.
i
1
.
pp
198
-
208
198
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
aas
.i
ae
s
c
or
e
.c
om
Im
p
r
ove
d
se
i
z
u
r
e
d
e
t
e
c
t
i
on
u
si
n
g o
p
t
i
m
i
z
e
d
t
i
m
e
se
q
u
e
n
c
e
b
ase
d
d
e
e
p
l
e
ar
n
i
n
g f
r
am
e
w
or
k
P
u
s
p
an
j
al
i
M
al
li
k
1,
2
, A
j
it
K
u
m
ar
N
ayak
3
, S
at
yap
r
ak
as
h
S
w
ai
n
4
1
D
e
pa
r
t
m
e
nt
of
C
om
put
e
r
S
c
i
e
nc
e
a
nd E
ngi
ne
e
r
i
ng, I
T
E
R
, S
O
A
D
e
e
m
e
d t
o be
U
ni
ve
r
s
i
t
y, O
di
s
ha
, I
ndi
a
2
S
ha
i
l
a
ba
l
a
W
om
e
n’
s
A
ut
onom
ous
C
ol
l
e
g
e
, C
ut
t
a
c
k
,
I
ndi
a
3
D
e
pa
r
t
m
e
nt
of
C
om
put
e
r
S
c
i
e
nc
e
a
nd I
nf
or
m
a
t
i
on T
e
c
hnol
ogy, I
T
E
R
, S
O
A
D
e
e
m
e
d t
o be
U
ni
ve
r
s
i
t
y,
B
huba
ne
s
w
a
r
,
I
ndi
a
4
D
e
pa
r
t
m
e
nt
of
C
om
put
e
r
S
c
i
e
nc
e
a
nd E
ngi
ne
e
r
i
ng, I
ns
t
i
t
ut
e
of
M
a
na
ge
m
e
nt
a
nd I
nf
or
m
a
t
i
on
T
e
c
hnol
ogy (
I
M
I
T
)
, C
ut
t
a
c
k, B
P
U
T
,
O
di
s
ha
, I
ndi
a
A
r
t
ic
le
I
n
f
o
A
B
S
T
R
A
C
T
A
r
ti
c
le
h
is
to
r
y
:
R
e
c
e
iv
e
d
J
un 7, 2025
R
e
vi
s
e
d
N
ov 18, 2025
A
c
c
e
pt
e
d
J
a
n 1, 2026
Epilepsy
disease
originates
due
to
the
presence
of
disordered
neuron
s
,
and
epilepsy
detection
stands
as
a
challengi
ng
task
for
neurolog
ists
.
With
recent
advances,
electroenceph
alography
(
EEG
)
-
based
analysis
is
increa
singly
supported
by
deep
learning
and
metaheuristic
optimization
approache
s
in
order
to
improve
the
test
results
.
This
experiment
uses
a
convol
utional
neural
network
(CNN)
model
hybridized
with
bidirectio
nal
long
shor
t
-
term
memory
(BiLSTM)
.
CNN
leverages
the
work
with
improved
f
eature
extractio
n
cum
classifi
cation
support
s
,
and
BiLSTM
keeps
the
time
sequence
of
data
in
both
the
forward
and
backward
direction
for
imp
roving
signal
mapping
purposes
.
To
reduce
the
computational
overhead
and
improve
execution
accura
cy,
a
hybrid
optimization
algorithm
called
secretar
y
bird
optimization
algorithm
(SBOA)
is
used
to
fine
-
tune
the
execution
.
Key
classification
parameters
such
as
accuracy,
sensitivity,
and
specificity
re
flect
the
model’s
strong
predictive
capability
,
with
ac
curacy
reaching
up
to
98.49%
.
The
proposed
method
demonstrates
the
potent
ial
for
high
-
performance
EEG
-
based
seizure
detection,
paving
the
way
for
future
integration
with
edge
computing
devices
to
support
remote
c
linical
diagnostics and continuous monitoring in real
-
world healthcare applicatio
ns.
K
e
y
w
o
r
d
s
:
B
iL
S
T
M
C
onvolut
io
na
l
ne
ur
a
l
ne
twor
k
E
le
c
tr
oe
nc
e
pha
lo
gr
a
m
M
e
ta
he
ur
is
ti
c
opt
im
iz
a
ti
on
S
e
c
r
e
ta
r
y bi
r
d opti
m
iz
a
ti
on
a
lg
or
it
hm
This is an
open
acce
ss artic
le unde
r the
CC BY
-
SA
license.
C
or
r
e
s
pon
di
n
g A
u
th
or
:
S
a
ty
a
pr
a
ka
s
h S
w
a
in
D
e
pa
r
tm
e
nt
of
C
om
put
e
r
S
c
ie
nc
e
a
nd E
ngi
ne
e
r
in
g
I
ns
ti
tu
te
of
M
a
na
ge
m
e
nt
a
nd I
nf
or
m
a
ti
on T
e
c
hnol
ogy (
I
M
I
T
)
,
C
ut
ta
c
k, B
P
U
T
O
di
s
ha
,
C
ut
ta
c
k,
I
ndi
a
E
m
a
il
:
s
a
ty
a
im
it
@
gm
a
il
.c
om
1.
I
N
T
R
O
D
U
C
T
I
O
N
A
s
e
pi
le
ps
y
is
a
li
f
e
-
di
s
tr
e
s
s
in
g
di
s
e
a
s
e
,
th
e
m
os
t
a
dvi
s
a
bl
e
s
te
p
is
to
de
te
c
t
th
e
pr
e
s
e
nc
e
of
s
e
iz
ur
e
s
ig
na
ls
in
or
de
r
to
pr
ovi
de
c
li
ni
c
a
l
s
ugge
s
ti
ons
to
s
a
ve
th
e
li
v
e
s
of
th
e
ne
ur
o
-
di
s
or
de
r
pa
ti
e
nt
s
.
P
r
e
s
e
nt
ly
,
a
c
onvolut
io
na
l
ne
ur
a
l
ne
twor
k
(
C
N
N
)
is
c
ons
id
e
r
e
d
a
s
th
e
a
dv
a
nc
e
d
c
la
s
s
if
ic
a
ti
on
te
c
hni
que
in
th
e
f
ie
ld
of
e
pi
le
ps
y
de
te
c
ti
on
,
w
hi
c
h
c
ons
i
s
ts
of
one
in
put
la
ye
r
,
one
out
put
la
ye
r
,
a
nd
m
or
e
th
a
n
one
c
onvolut
io
na
l
la
ye
r
,
w
he
r
e
th
e
in
put
la
ye
r
is
c
onne
c
te
d
w
it
h
th
e
f
ir
s
t
c
onvolu
ti
ona
l
la
ye
r
a
nd
th
e
out
put
la
ye
r
is
c
onne
c
te
d
to
th
e
la
s
t
c
onvolut
io
na
l
la
ye
r
[
1]
.
T
hi
s
C
N
N
c
om
pr
is
e
s
m
il
li
ons
of
ne
ur
ons
,
a
nd
e
a
c
h
ne
ur
on
is
m
a
th
e
m
a
ti
c
a
ll
y
e
xpr
e
s
s
e
d
a
s
:
=
(
+
)
,
‘
y’
is
th
e
out
put
pa
r
a
m
e
te
r
,
‘
x’
is
th
e
in
put
pa
r
a
m
e
te
r
,
‘
W
’
is
th
e
w
e
ig
ht
m
a
tr
ix
,
‘
b
’
is
th
e
bi
a
s
va
lu
e
,
a
nd
‘
f
’
is
th
e
a
c
ti
va
ti
o
n
f
unc
ti
on
[
2
]
.
A
ls
o,
m
e
ta
he
ur
is
ti
c
a
lg
or
it
hm
s
a
r
e
e
m
e
r
gi
ng
a
s
th
e
m
o
s
t
pow
e
r
f
ul
a
nd
popula
r
ly
us
e
d
a
lg
or
it
hm
s
,
w
hi
c
h
a
r
e
in
te
gr
a
te
d
w
it
h
m
a
c
hi
n
e
le
a
r
ni
ng
a
lg
or
it
hm
s
[
3
]
.
T
he
r
e
a
r
e
di
f
f
e
r
e
nt
ty
pe
s
of
m
e
ta
h
e
ur
is
ti
c
a
lg
or
it
hm
s
,
s
uc
h
a
s
na
tu
r
e
-
in
s
pi
r
e
d,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
dv A
ppl
S
c
i
I
S
S
N
:
2252
-
8814
I
m
pr
ov
e
d s
e
i
z
ur
e
d
e
te
c
ti
on us
in
g opti
m
iz
e
d t
ime
s
e
qu
e
nc
e
ba
s
e
d de
e
p l
e
ar
ni
ng
…
(
P
us
panj
al
i
M
al
li
k
)
199
bi
o
-
in
s
pi
r
e
d,
s
w
a
r
m
-
in
s
pi
r
e
d
,
a
nd
tr
a
je
c
to
r
y
-
ba
s
e
d
a
lg
or
it
h
m
s
[
4]
.
T
he
r
e
a
r
e
di
f
f
e
r
e
nt
ty
pe
s
of
ne
ur
a
l
di
s
or
de
r
s
a
nd
P
a
r
ki
ns
on’
s
di
s
e
a
s
e
is
one
a
m
ong
th
e
m
c
ha
r
a
c
t
e
r
iz
e
d
by
s
ho
w
in
g
s
lo
w
m
ove
m
e
nt
,
pos
tu
r
a
l
in
s
ta
bi
li
ty
, r
ig
id
it
y
r
e
f
le
c
ti
ng t
he
de
ge
ne
r
a
ti
ve
c
ondi
ti
on of
t
he
br
a
in
.
A
lt
hough the
t
r
e
a
tm
e
nt
of
t
hi
s
di
s
e
a
s
e
i
s
a
ti
m
e
s
pa
n
f
a
c
to
r
,
s
ti
ll
th
e
n
th
e
pe
r
f
e
c
t
a
na
ly
s
i
s
of
th
e
di
s
e
a
s
e
w
il
l
be
c
ons
id
e
r
e
d
a
s
a
c
or
e
p
a
r
t
of
c
li
ni
c
a
l
pr
oc
e
dur
e
s
.
T
il
l
ye
t,
th
e
e
le
c
tr
oe
nc
e
pha
lo
gr
a
phy
(
EEG
)
te
s
t
a
nd
m
a
gne
ti
c
r
e
s
ona
nc
e
im
a
gi
ng
(
M
R
I
)
of
th
e
s
c
a
lp
a
r
e
us
e
f
ul
to
di
a
gnos
e
th
e
s
ym
pt
om
s
of
P
a
r
ki
ns
on’
s
di
s
e
a
s
e
.
I
t
r
e
qui
r
e
s
th
e
c
or
r
e
c
t
s
e
iz
ur
e
s
ig
na
l
a
na
ly
s
is
,
a
nd
pr
e
f
e
r
a
bl
y
,
th
e
be
s
t
a
n
a
ly
s
is
i
s
pe
r
f
or
m
e
d
w
it
h
de
e
p
le
a
r
ni
ng
a
lg
or
it
hm
s
li
ke
s
ub
-
ba
nd
a
na
ly
s
is
w
it
h
ga
te
d
r
e
c
ur
r
e
nt
uni
t
(
G
R
U
)
[
5]
a
nd
m
ul
ti
s
c
a
le
C
N
N
[
6]
.
S
om
e
ti
m
e
s
,
th
e
r
e
a
r
e
a
ls
o
th
e
pos
s
ib
il
it
ie
s
of
a
r
ti
f
a
c
ts
in
te
r
m
ix
ed
w
it
h
th
e
s
ig
na
ls
,
due
to
w
hi
c
h
th
e
c
or
r
e
c
t
s
e
iz
ur
e
pr
e
di
c
ti
on
a
nd
de
te
c
ti
on
b
e
c
om
e
to
ugh
,
a
nd
th
e
pe
r
f
or
m
a
nc
e
a
c
c
ur
a
c
y
be
c
om
e
s
lo
w
e
r
.
I
n
th
is
c
a
s
e
,
th
e
de
e
p
le
a
r
ni
ng
m
ode
ls
m
us
t
be
in
te
gr
a
te
d
w
it
h
le
a
r
na
bl
e
c
a
p
a
c
it
y
,
a
nd
th
e
m
od
e
l
is
c
a
ll
e
d
a
l
e
a
r
na
bl
e
a
nd
e
xpl
a
in
a
bl
e
w
a
v
e
le
t
ne
ur
a
l
ne
twor
k
[
7]
,
[
8]
.
W
he
n
th
e
E
E
G
s
ig
na
l
is
r
e
c
or
de
d
,
it
a
ppe
a
r
s
ir
r
e
gul
a
r
a
nd
non
-
s
m
o
ot
h
f
or
w
hi
c
h
i
t
be
c
om
e
s
di
f
f
ic
ul
t
to
m
e
a
s
ur
e
s
e
iz
ur
e
f
r
e
que
nc
y
[
9]
.
A
li
ght
w
e
ig
ht
c
onvolut
io
n t
r
a
ns
f
or
m
e
r
(
L
C
T
)
i
s
pr
opos
e
d
[
10]
f
or
c
r
os
s
-
pa
ti
e
nt
s
e
iz
ur
e
de
te
c
ti
on
,
w
hi
c
h pr
oduc
e
s
s
m
oot
hne
s
s
i
n s
e
iz
ur
e
s
ig
na
ls
.
D
e
pr
e
s
s
io
n
is
a
s
pe
c
ia
l
c
a
s
e
of
a
s
e
iz
ur
e
s
ym
pt
om
th
a
t
m
a
y
h
a
ppe
n
due
to
pe
r
s
ona
l
pr
obl
e
m
s
,
a
nd
de
e
p
de
pr
e
s
s
io
n
a
f
f
e
c
ts
th
e
f
r
e
e
th
in
ki
ng
o
f
a
pe
r
s
on
.
T
he
us
e
f
ul
to
ol
s
e
m
pl
oye
d
f
o
r
de
pr
e
s
s
io
n
de
te
c
ti
on
a
r
e
a
n
e
nc
ode
r
f
or
da
ta
c
om
pr
e
s
s
io
n,
lo
ng
s
hor
t
-
te
r
m
m
e
m
or
y
(
L
S
T
M
)
f
or
e
xpr
e
s
s
in
g
th
e
te
m
por
a
l
vi
br
a
ti
ons
of
th
e
r
e
c
or
de
d
s
ig
na
l
,
a
nd
a
n
a
tt
e
nt
io
n
m
e
c
ha
ni
s
m
to
in
tr
oduc
e
pa
r
a
ll
e
li
s
m
a
m
ong
th
e
c
om
pr
e
s
s
e
d
in
f
or
m
a
ti
on
[
11]
.
B
ut
i
f
th
e
s
ig
na
l
is
c
om
pl
e
x
a
nd
no
is
y
in
na
tu
r
e
,
th
e
n
th
e
R
ie
m
a
nni
a
n
s
pe
c
tr
a
l
c
lu
s
te
r
in
g
[
12]
m
e
th
od
is
f
ol
lo
w
e
d
to
id
e
nt
if
y
th
e
out
li
e
r
s
[
13
]
.
T
he
r
e
a
r
e
s
om
e
s
pe
c
ia
l
c
a
s
e
s
w
he
r
e
da
ta
s
e
t
pr
iv
a
c
y i
s
m
a
in
ta
in
e
d
[
14]
a
lo
ng
w
it
h s
e
iz
ur
e
de
te
c
ti
on f
r
o
m
t
h
e
pr
iv
a
c
y point
of
vi
e
w
of
pa
ti
e
nt
i
nf
or
m
a
ti
on
.
S
om
e
ty
pi
c
a
l
ne
ur
a
l
n
e
twor
ks
a
r
e
de
vi
s
e
d
f
or
s
pe
c
if
ic
di
s
e
a
s
e
s
,
li
ke
A
lz
he
im
e
r
’
s
d
is
e
a
s
e
de
te
c
ti
on
[
15]
pe
r
f
or
m
e
d
w
it
h
A
da
z
d
-
N
e
t
[
16]
a
nd
a
n
a
ut
om
a
te
d
de
e
p
ne
ur
a
l
ne
twor
k
[
17]
m
ode
l
.
W
a
ng
e
t
al
.
[
18
]
pr
opos
e
d
a
hybr
id
m
ode
l
us
in
g
a
s
uppor
t
ve
c
to
r
m
a
c
hi
n
e
(
S
V
M
)
a
nd
ke
r
ne
l
s
pa
r
s
e
r
e
pr
e
s
e
nt
a
ti
on
c
la
s
s
if
ic
a
ti
on
(
K
S
R
C
)
;
W
u
e
t
al
.
[
19]
pr
opos
e
d
a
s
pa
ti
a
l
f
e
a
tu
r
e
f
us
e
d
c
onvolut
io
na
l
ne
twor
k
(
S
C
N
e
t)
f
o
r
E
E
G
pa
th
ol
ogy de
te
c
ti
on
.
T
o
ove
r
c
om
e
th
e
pr
oc
e
s
s
in
g
c
om
pl
e
xi
ty
,
m
e
ta
he
ur
is
ti
c
opt
im
iz
a
ti
on
a
lg
or
it
hm
s
a
r
e
in
te
gr
a
te
d
w
it
h
ne
ur
a
l
ne
twor
ks
.
T
hi
s
pa
pe
r
us
e
s
th
e
s
e
c
r
e
ta
r
y
bi
r
d
opt
im
iz
a
ti
on
a
lg
or
it
hm
(
S
B
O
A
)
to
f
in
e
-
tu
ne
th
e
m
ode
l
a
nd
to
r
e
duc
e
th
e
ope
r
a
ti
ona
l
ove
r
he
a
d
[
20]
.
D
e
e
p
le
a
r
ni
ng
w
it
h
s
e
que
nt
ia
l
a
r
r
a
nge
m
e
nt
[
21]
in
te
gr
a
ti
ng
w
it
h
L
S
T
M
[
22]
e
ns
ur
e
s
da
t
a
de
pe
nde
n
c
ie
s
ov
e
r
ti
m
e
,
a
nd
a
l
s
o
r
e
a
l
-
time
-
ba
s
e
d
de
e
p
le
a
r
ni
ng
m
ode
ls
[
23]
,
[
24]
e
ns
ur
e
E
E
G
de
te
c
ti
on
in
a
c
r
uc
ia
l
ti
m
e
f
r
a
m
e
[
25]
.
M
os
t
of
th
e
w
or
k
im
pl
e
m
e
nt
s
E
E
G
de
te
c
ti
on
by
in
te
gr
a
ti
ng
de
e
p
le
a
r
ni
ng
a
nd
a
b
id
ir
e
c
ti
ona
l
lo
ng
s
hor
t
-
te
r
m
m
e
m
or
y
(
B
iL
S
T
M
)
m
ode
l
to
a
na
ly
z
e
s
pa
ti
a
l
r
e
la
ti
ons
hi
p a
m
ong
E
E
G
s
ig
na
l
by C
N
N
,
a
nd t
he
n t
e
m
por
a
l
a
na
ly
s
is
by us
in
g B
iL
S
T
M
[
26]
–
[
28]
.
A
ll
th
e
m
e
nt
io
ne
d
r
e
vi
e
w
a
r
ti
c
le
s
e
nh
a
nc
e
th
e
a
c
c
ur
a
te
E
E
G
de
te
c
ti
on
by
in
te
gr
a
ti
ng
opt
im
iz
a
ti
on
te
c
hni
que
s
,
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
m
e
th
ods
,
a
nd
s
ta
ti
s
ti
c
a
l
a
ppr
oa
c
he
s
to
m
a
c
hi
ne
in
te
ll
ig
e
nc
e
.
A
lt
hough
no
t
a
bl
e
im
pr
ove
m
e
nt
s
a
r
e
a
lr
e
a
dy
pl
a
c
e
d
in
s
e
iz
ur
e
de
t
e
c
ti
on,
s
e
v
e
r
a
l
s
ig
ni
f
ic
a
nt
li
m
it
a
ti
ons
s
ti
ll
e
xi
s
t
.
N
um
e
r
ous
te
c
hni
que
s
in
vol
ve
in
tr
ic
a
te
pr
e
pr
oc
e
s
s
in
g,
w
hi
c
h
ha
m
pe
r
s
th
e
ir
pr
a
c
ti
c
a
l
us
e
in
r
e
a
l
-
ti
m
e
s
y
s
te
m
s
.
S
om
e
m
ode
ls
a
r
e
t
e
s
te
d only on
s
pe
c
if
ic
da
ta
s
e
ts
, l
im
it
in
g t
he
ir
ge
ne
r
a
li
z
a
bi
li
ty
a
c
r
os
s
di
f
f
e
r
e
nt
E
E
G
s
ig
na
ls
.
T
o
a
ddr
e
s
s
th
e
li
m
it
a
ti
ons
of
e
a
r
li
e
r
s
e
iz
ur
e
de
te
c
ti
on
m
ode
ls
,
w
e
pr
opos
e
a
n
e
nha
nc
e
d
a
ppr
oa
c
h
th
a
t
c
om
bi
ne
s
a
C
N
N
a
nd
B
iL
S
T
M
to
pe
r
f
or
m
th
e
r
obus
t
f
e
a
t
ur
e
e
xt
r
a
c
ti
on,
c
la
s
s
if
ic
a
ti
on
,
a
nd
p
e
r
ta
in
in
g
of
ti
m
e
s
e
r
ie
s
da
ta
e
le
m
e
nt
s
.
F
ur
th
e
r
,
it
us
e
s
a
n
im
pr
ove
d
m
e
ta
he
ur
is
ti
c
opt
im
iz
a
ti
on
te
c
hni
que
,
S
B
O
A
,
to
f
in
e
-
tu
ne
th
e
m
ode
l
a
nd
to
r
e
duc
e
th
e
pr
oc
e
s
s
in
g
c
om
pl
e
xi
ty
.
T
hi
s
hybr
id
m
e
th
od
f
in
e
-
tu
ne
s
th
e
m
ode
l
pa
r
a
m
e
te
r
s
e
f
f
ic
ie
nt
ly
,
r
e
duc
in
g
tr
a
in
in
g
lo
s
s
a
nd
ti
m
e
.
O
ur
f
i
na
l
r
e
s
ul
t,
a
c
hi
e
ve
d
a
t
100
e
poc
hs
,
s
how
s
a
n
a
c
c
ur
a
c
y of
98.49%
, s
e
ns
it
iv
it
y of
96.05%
,
s
pe
c
if
ic
it
y of
97.03
%
,
M
a
tt
he
w
s
c
or
r
e
la
ti
on c
oe
f
f
ic
ie
nt
(
M
C
C
)
of
97.01%
, a
nd
a
r
e
a
unde
r
t
he
c
ur
ve
(
AUC
)
of
0.97.
T
he
r
e
s
t
of
our
a
r
ti
c
le
is
e
la
bor
a
te
d
in
s
e
c
ti
on
s
2
to
4
.
S
e
c
ti
on
2
is
th
e
c
or
e
pa
r
t
of
th
e
e
xpe
r
im
e
n
t
e
xpl
a
in
s
a
bout
th
e
m
e
th
od
us
e
d
w
it
h
th
e
s
ub
-
s
e
c
ti
ons
of
2.1
th
a
t
e
la
bor
a
te
s
f
lo
w
of
w
or
k,
2.2
e
la
bor
a
te
s
th
e
f
e
a
tu
r
e
s
of
e
xpe
r
im
e
nt
a
l
B
onn
E
E
G
da
ta
s
e
t,
2.3
s
ta
te
s
th
e
te
c
h
ni
que
s
of
E
E
G
da
ta
pr
e
-
p
r
oc
e
s
s
in
g,
2.4
s
ta
te
s
da
ta
a
ugm
e
nt
a
ti
on,
2.5
s
ta
te
s
a
bout
th
e
vot
in
g
m
ode
l
s
,
2.6
pr
e
s
e
nt
s
th
e
B
iL
S
T
M
ne
twor
k
,
a
nd
2.7
pr
e
s
e
nt
s
th
e
opt
im
iz
e
d
S
B
O
A
a
lg
or
it
hm
.
S
e
c
ti
on
3
pr
e
s
e
nt
s
e
xpe
r
im
e
nt
a
l
r
e
s
ul
ts
a
nd
di
s
c
us
s
io
ns
,
a
nd
s
e
c
ti
on
4
s
ta
te
s
th
e
c
onc
lu
s
io
n a
nd f
ut
ur
e
s
c
ope
of
t
he
p
a
pe
r
.
2.
M
E
T
H
O
D
2.1
.
F
lo
w
of
w
or
k
T
hi
s
pr
opos
e
d
m
ode
l
f
or
E
E
G
de
te
c
ti
on
,
de
pi
c
te
d
in
F
ig
ur
e
1
,
is
a
f
r
a
m
e
w
or
k
to
opt
im
iz
e
th
e
te
s
t
a
c
c
ur
a
c
y
a
nd
to
r
e
duc
e
th
e
pr
oc
e
s
s
in
g
c
om
pl
e
xi
ty
.
A
t
f
ir
s
t,
i
t
a
ppl
ie
s
th
e
s
e
que
nc
e
of
th
e
pr
e
-
p
r
oc
e
s
s
in
g
pi
pe
li
ne
to
im
pr
ove
da
ta
qua
li
ty
,
f
ol
lo
w
e
d
by
da
ta
a
ugm
e
nt
a
ti
on
f
or
im
pr
ovi
ng
th
e
m
ode
l’
s
ge
ne
r
a
li
z
a
ti
on
.
T
he
n i
t
tr
ie
s
f
or
de
e
p l
e
a
r
ni
ng
m
ode
ls
f
or
c
la
s
s
if
ic
a
ti
on.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8814
I
nt
J
A
dv A
ppl
S
c
i
,
V
ol
.
15, No
.
1, M
a
r
c
h 2026
:
198
-
208
200
2.2. Clin
ic
al
d
at
as
e
t
s
T
h
e
B
on
n
U
ni
ve
r
s
it
y
d
a
t
a
s
e
t
,
w
hi
c
h
i
s
c
o
ll
e
c
t
e
d
f
r
om
P
hy
s
io
n
e
t
,
i
s
a
pu
bl
i
c
l
y
a
v
a
i
la
bl
e
E
E
G
da
ta
ba
s
e
c
e
nt
e
r
[
2
1]
.
I
t
i
s
a
m
u
lt
i
c
l
a
s
s
E
E
G
s
i
gn
a
l
da
ta
ba
s
e
.
T
he
r
e
a
r
e
f
iv
e
c
l
a
s
s
e
s
pr
e
s
e
n
t
in
th
e
d
a
t
a
s
e
t
s
,
c
l
a
s
s
A
t
o
c
l
a
s
s
E
,
a
nd
e
a
c
h
d
a
t
a
s
e
t
c
o
nt
a
in
s
100
tx
t
f
il
e
s
w
i
th
4
,
0
96
s
a
m
pl
e
s
i
n
A
S
C
I
I
f
or
m
a
t
.
C
la
s
se
s
A
t
o
D
c
o
nt
a
in
nor
m
a
l
s
ig
n
a
l
a
nd
c
l
a
s
s
E
c
o
nt
a
in
s
s
e
i
z
ur
e
s
i
gn
a
l
s
.
T
h
e
r
e
a
r
e
1
00
c
ha
nn
e
l
s
p
r
e
s
e
nt
in
e
a
c
h
da
ta
s
e
t
.
F
or
c
ol
le
c
ti
ng
E
E
G
s
i
gn
a
l,
t
he
e
le
c
tr
od
e
s
a
r
e
ke
pt
o
ve
r
t
he
h
e
a
d
of
t
he
p
a
ti
e
n
t
a
n
d
in
s
i
de
i
nt
r
a
c
r
a
n
ia
l
r
e
g
io
n
s
of
t
he
he
a
d
f
or
a
ti
m
e
pe
r
io
d o
f
2
3.
6
s
e
c
o
nd
s
.
F
i
gur
e
2 d
e
p
ic
t
s
t
he
s
a
m
p
le
im
a
g
e
s
c
ol
l
e
c
t
e
d
f
r
om
E
E
G
p
a
ti
e
n
ts
.
F
ig
ur
e
1
.
P
r
opos
e
d m
ode
l
de
s
c
r
ib
in
g t
he
f
lo
w
of
t
he
w
or
k
F
ig
ur
e
2
.
B
onn mul
ti
c
la
s
s
E
E
G
s
ig
na
l
im
a
ge
s
2.3
.
D
at
a p
r
e
-
p
r
oc
e
s
s
in
g
D
a
ta
pr
e
-
pr
oc
e
s
s
in
g
is
a
pi
pe
li
ne
of
c
onve
r
s
io
n
a
nd
f
il
tr
a
ti
on
te
c
hni
que
s
th
a
t
c
om
e
s
a
f
te
r
da
ta
c
ol
le
c
ti
on
to
boos
t
th
e
a
c
c
ur
a
c
y
a
nd
c
ons
is
te
nc
y
of
th
e
te
s
t
r
e
s
ul
t
.
I
ni
ti
a
ll
y,
th
e
im
a
ge
s
w
e
r
e
c
onve
r
te
d
to
gr
a
ys
c
a
le
,
f
ol
lo
w
e
d
by
noi
s
e
r
e
duc
ti
on
us
in
g
G
a
us
s
ia
n
a
nd
m
e
di
a
n
f
il
te
r
in
g
.
C
ont
r
a
s
t
w
a
s
th
e
n
e
nha
nc
e
d
us
in
g
c
ont
r
a
s
t
li
m
it
e
d
a
dopt
iv
e
hi
s
to
gr
a
m
e
qua
li
z
a
ti
on
(
C
L
A
H
E
)
to
hi
gh
li
ght
im
po
r
ta
nt
f
e
a
tu
r
e
s
.
T
he
im
a
ge
s
w
e
r
e
r
e
s
iz
e
d
to
a
s
ta
nda
r
d
di
m
e
ns
io
n
,
a
nd
min
-
m
a
x
nor
m
a
li
z
a
ti
on
w
a
s
a
ppl
ie
d
to
s
c
a
le
pi
xe
l
v
a
lu
e
s
uni
f
or
m
ly
, e
ns
ur
in
g r
e
a
di
ne
s
s
f
or
a
c
c
ur
a
te
a
nd e
f
f
ic
ie
nt
a
na
ly
s
i
s
.
F
ig
ur
e
3 s
how
s
t
he
pr
e
-
pr
oc
e
s
s
e
d i
m
a
ge
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
dv A
ppl
S
c
i
I
S
S
N
:
2252
-
8814
I
m
pr
ov
e
d s
e
i
z
ur
e
d
e
te
c
ti
on us
in
g opti
m
iz
e
d t
ime
s
e
qu
e
nc
e
ba
s
e
d de
e
p l
e
ar
ni
ng
…
(
P
us
panj
al
i
M
al
li
k
)
201
2.4. Dat
a au
gm
e
n
t
at
io
n
T
o
e
nha
nc
e
th
e
va
r
ia
bi
li
ty
of
E
E
G
im
a
ge
s
a
m
pl
e
s
a
nd
a
m
e
li
or
a
te
th
e
m
ode
l’
s
ge
ne
r
a
li
z
a
ti
on
c
a
pa
bi
li
ty
,
a
c
om
pr
e
he
ns
iv
e
da
ta
a
ugm
e
nt
a
ti
on
s
tr
a
te
gy
w
a
s
e
m
pl
oye
d.
T
hi
s
pr
oc
e
s
s
he
lp
s
to
m
it
ig
a
te
ove
r
f
it
ti
ng
by
in
tr
oduc
in
g
di
ve
r
s
e
vi
s
ua
l
pa
tt
e
r
ns
in
to
th
e
tr
a
in
in
g
da
ta
s
e
t.
T
h
e
a
ppl
ie
d
tr
a
ns
f
or
m
a
ti
ons
in
c
lu
de
d
im
a
ge
r
ot
a
ti
on,
hor
iz
ont
a
l
a
nd
ve
r
ti
c
a
l
f
li
ppi
ng,
r
e
s
i
z
in
g,
c
ol
or
ji
tt
e
r
in
g,
th
e
a
ddi
ti
on
of
G
a
us
s
ia
n
noi
s
e
, a
nd i
nt
e
ns
it
y m
odi
f
ic
a
ti
ons
. T
he
s
e
a
ugm
e
nt
a
ti
ons
s
im
ul
a
te
r
e
a
li
s
ti
c
a
lt
e
r
a
ti
ons
t
ha
t
c
oul
d oc
c
ur
i
n E
E
G
im
a
gi
ng
c
ondi
ti
ons
,
th
e
r
e
by
e
na
bl
in
g
th
e
m
ode
l
to
le
a
r
n
m
or
e
r
obus
t
a
nd
ge
ne
r
a
li
z
e
d
r
e
pr
e
s
e
nt
a
ti
ons
.
E
xa
m
pl
e
s
of
t
he
s
e
a
ugm
e
nt
e
d i
m
a
g
e
s
a
r
e
i
ll
us
tr
a
te
d i
n F
ig
ur
e
4
.
F
ig
ur
e
3
.
P
r
e
-
pr
oc
e
s
s
e
d i
m
a
ge
of
E
E
G
s
ig
na
l
F
ig
ur
e
4
.
D
a
ta
a
ugm
e
nt
a
ti
on i
m
a
ge
s
2.5
.
V
ot
in
g m
od
e
l
A
f
te
r
th
e
da
ta
a
ugm
e
nt
a
ti
on
pha
s
e
,
th
e
e
ns
e
m
bl
e
le
a
r
ni
ng
m
e
t
hod
is
a
ppl
ie
d
to
th
e
s
e
i
z
ur
e
s
ig
na
l
to
ge
t
th
e
opt
im
um
r
e
s
ul
t
.
H
e
r
e
,
th
e
e
n
s
e
m
bl
e
le
a
r
ni
ng
te
c
hni
q
ue
us
e
s
D
-
N
e
t,
R
e
s
N
e
t,
G
-
N
e
t
,
a
nd
C
N
N
a
s
th
e
vot
in
g
c
la
s
s
if
ie
r
,
a
nd
th
e
r
e
s
ul
t
in
g
out
put
is
th
e
c
om
bi
ne
d
m
a
jo
r
it
y
of
e
a
c
h
of
th
e
vo
ti
ng
c
la
s
s
if
ie
r
s
.
T
he
1D
-
C
N
N
us
e
d
he
r
e
i
s
a
s
e
que
n
ti
a
l
C
N
N
bui
l
t
up
w
it
h
la
ye
r
s
s
u
c
h
a
s
C
onv1D, M
a
xP
ool
in
g1D
,
f
la
tt
e
n,
de
ns
e
,
a
nd
dr
opout
la
ye
r
s
.
T
he
C
onv1D
la
ye
r
u
s
e
s
64
f
il
te
r
s
,
ke
r
ne
l
s
iz
e
is
3,
a
c
ti
va
ti
on
f
unc
ti
on
us
e
d
is
“
r
e
c
ti
f
ie
d
li
ne
a
r
uni
t
(
R
e
L
U
)
,
”
a
nd
it
pr
oc
e
s
s
e
s
E
E
G
s
ig
na
ls
in
s
e
qu
e
nt
ia
l
or
de
r
.
I
n
th
e
M
a
xP
ool
in
g1D
la
ye
r
,
C
N
N
us
e
s
a
w
in
dow
of
pool
s
iz
e
2
.
T
he
f
la
tt
e
n
ed
l
a
ye
r
r
e
-
a
r
r
a
nge
s
th
e
e
xt
r
a
c
te
d
f
e
a
tu
r
e
s
a
nd
th
e
n
f
la
tt
e
ns
th
e
c
ha
nne
l
di
m
e
ns
io
n
.
T
he
de
ns
e
la
ye
r
u
s
e
s
256
ne
ur
on
s
,
a
nd
th
e
a
c
t
iv
a
ti
on
f
unc
ti
on
us
e
d
he
r
e
is
R
e
L
U
.
R
e
L
U
pr
oc
e
s
se
s
th
e
s
ig
na
ls
a
nd
h
e
lp
s
th
e
C
N
N
to
le
a
r
n
nonl
in
e
a
r
pa
tt
e
r
ns
.
T
he
d
r
opout
la
ye
r
us
e
s
r
e
gul
a
r
iz
a
ti
on
te
c
hni
que
f
or
pr
e
ve
nt
in
g
ove
r
f
it
ti
ng
in
C
N
N
.
I
t
ha
s
a
dr
opout
r
a
te
of
0.5
.
T
hi
s
C
N
N
us
e
s
de
n
s
e
la
ye
r
s
a
nd
dr
opout
la
ye
r
a
lt
e
r
na
ti
ve
ly
.
T
he
r
e
s
ul
t
is
s
pe
c
if
ie
d
in
T
a
bl
e
1
.
F
ig
ur
e
5
de
pi
c
ts
th
e
s
te
ps
of
s
e
le
c
ti
ng
th
e
id
e
a
l
vot
in
g
m
ode
l
a
nd
T
a
bl
e
1 s
to
r
e
s
t
he
a
c
c
ur
a
c
y, l
os
s
a
nd va
l
-
lo
s
s
va
lu
e
s
of
t
he
c
hos
e
n voti
ng mode
ls
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8814
I
nt
J
A
dv A
ppl
S
c
i
,
V
ol
.
15, No
.
1, M
a
r
c
h 2026
:
198
-
208
202
T
a
bl
e
1
.
V
ot
in
g
c
la
s
s
if
ie
r
pe
r
f
or
m
a
nc
e
r
e
s
ul
t
L
a
ye
r
O
ut
put
C
onv1D
(
18, 64)
M
a
xP
ool
i
ng1D
(
9, 64)
C
onv1D
(
7, 128)
M
a
xP
ool
i
ng1D
(
3, 128)
F
l
a
t
t
e
n
(
384)
D
e
ns
e
(
256)
D
r
opout
(
256)
D
e
ns
e
(
128)
D
r
opout
(
128)
D
e
ns
e
(
1)
F
ig
ur
e
5
.
S
te
ps
of
c
hoos
in
g be
s
t
vot
in
g m
ode
l
a
s
C
N
N
S
uc
c
e
s
s
f
ul
ly
r
un
th
e
m
ode
ls
li
s
te
d
in
T
a
bl
e
2
,
a
nd
a
f
te
r
c
o
m
pl
e
ti
on
,
it
is
f
ound
th
a
t
th
e
r
e
s
ul
t
obt
a
in
e
d
in
C
N
N
m
ode
l
s
upe
r
s
e
d
e
s
th
e
r
e
s
ul
t
s
w
it
h
D
-
N
e
t,
R
e
s
N
e
t
,
a
nd
G
-
N
e
t.
T
he
a
c
c
ur
a
te
r
a
te
of
C
N
N
is
89.08,
lo
s
s
va
lu
e
is
0.16
,
a
nd
v
al
-
lo
s
s
va
lu
e
is
0.15.
T
he
n,
te
s
t
th
e
ot
he
r
pe
r
f
or
m
a
nc
e
m
e
tr
ic
s
to
p
r
ove
th
e
s
upe
r
io
r
it
y
of
C
N
N
m
ode
l
a
m
ong
D
-
N
e
t,
R
e
s
N
e
t
,
a
nd
G
-
N
e
t
m
ode
ls
.
T
he
pe
r
f
or
m
a
nc
e
m
e
a
s
ur
e
s
in
te
r
m
s
of
a
c
c
ur
a
c
y, s
e
ns
it
iv
it
y, s
pe
c
if
ic
it
y
,
a
nd A
U
C
s
c
or
e
s
a
r
e
l
is
te
d i
n
T
a
bl
e
3.
T
a
bl
e
2
.
A
c
c
ur
a
c
y,
lo
s
s
,
a
nd va
l
-
lo
s
s
of
vot
in
g m
ode
ls
D
-
N
e
t
R
e
s
N
e
t
G
-
N
e
t
C
N
N
A
c
y
L
os
s
V
-
l
os
s
A
c
y
L
os
s
V
-
l
os
s
A
c
y
L
os
s
V
-
l
os
s
A
c
y
L
os
s
V
-
l
os
s
T
e
s
t
1
75.02
0.25
0.33
79.02
0.17
0.32
80.01
0.27
0.34
89.01
0.26
0.36
T
e
s
t
2
75.04
0.24
0.21
79.02
0.16
0.32
80.02
0.28
0.24
89.03
0.21
0.32
T
e
s
t
3
75.06
0.22
0.25
79.04
0.23
0.31
80.05
0.25
0.33
89.04
0.18
0.25
T
e
s
t
4
75.06
0.18
0.23
79.05
0.28
0.31
80.06
0.21
0.29
89.06
0.19
0.29
T
e
s
t
5
75.82
0.17
0.21
79.06
0.22
0.29
80.07
0.19
0.28
89.08
0.16
0.15
T
a
bl
e
3
.
P
e
r
f
or
m
a
nc
e
m
e
tr
ic
s
of
pr
e
f
e
r
r
e
d m
ode
ls
M
ode
l
s
A
c
c
ur
a
c
y (
%
)
S
e
ns
i
t
i
vi
t
y (
%
)
S
pe
c
i
f
i
c
i
t
y (
%
)
M
C
C
AUC
D
-
N
e
t
75.82
78.62
76.09
78.03
0.77
R
e
s
N
e
t
79.06
79.19
79.28
80.32
0.79
G
-
N
e
t
80.07
80.09
81.03
80.04
0.80
C
N
N
89.08
89.01
90.01
89.09
0.89
2.6
.
B
id
ir
e
c
t
io
n
al
lo
n
g s
h
or
t
-
t
e
r
m
m
e
m
or
y
A
f
t
e
r
t
h
e
s
i
gn
a
l
s
a
r
e
e
xt
r
a
c
t
e
d
by v
ot
i
ng
m
od
e
l
s
, i
t
r
e
q
ui
r
e
s
a
n a
p
pr
opr
ia
te
or
d
e
r
i
ng of
r
e
pr
e
s
e
n
ta
t
io
n
i
s
r
e
q
ui
r
e
d
.
H
e
r
e
,
th
e
e
x
tr
a
c
t
e
d
E
E
G
s
ig
na
ls
u
nd
e
r
w
e
n
t
t
hr
o
ugh
a
t
im
e
s
e
r
i
e
s
s
e
qu
e
nt
ia
l
r
e
pr
e
s
e
nt
a
ti
on
of
d
a
t
a
or
d
e
r
in
g,
c
a
ll
e
d
a
s
B
iL
S
T
M
,
w
hi
c
h
c
a
n
s
u
c
c
e
s
s
f
ul
ly
h
a
n
dl
e
p
e
r
s
i
s
t
e
nt
d
e
p
e
n
de
nc
ie
s
in
ti
m
e
s
e
qu
e
n
c
i
ng
da
t
a
[
2
6]
a
s
s
pe
c
if
ie
d
in
T
a
bl
e
4
,
a
nd
F
i
gur
e
6
d
e
pi
c
t
s
a
r
c
hi
t
e
c
t
ur
e
of
B
iL
S
T
M
m
o
de
l
.
B
iL
S
T
M
in
t
e
gr
a
t
e
s
th
e
s
i
gn
a
l
f
lo
w
of
t
w
o L
S
T
M
S
,
on
e
i
n f
or
w
a
r
d
d
ir
e
c
ti
on
a
nd
ot
he
r
i
n
r
e
ve
r
s
e
di
r
e
c
t
io
n
,
w
it
h
a
s
i
ng
le
ou
tp
u
t
l
a
y
e
r
.
T
a
bl
e
4
.
D
e
p
e
nde
nc
ie
s
i
n
ti
m
e
s
e
que
n
c
in
g da
ta
L
a
ye
r
O
ut
put
I
nput
(
178, 1)
D
e
ns
e
(
178, 32)
B
i
di
r
e
c
t
i
on
(
256)
D
r
opout
(
256)
B
a
t
c
h_nor
m
a
l
i
z
a
t
i
on
(
256)
D
e
ns
e
(
64)
D
r
opout
(
64)
B
a
t
c
h_nor
m
a
l
i
z
a
t
i
on
(
64)
D
e
ns
e
(
2)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
dv A
ppl
S
c
i
I
S
S
N
:
2252
-
8814
I
m
pr
ov
e
d s
e
i
z
ur
e
d
e
te
c
ti
on us
in
g opti
m
iz
e
d t
ime
s
e
qu
e
nc
e
ba
s
e
d de
e
p l
e
ar
ni
ng
…
(
P
us
panj
al
i
M
al
li
k
)
203
F
ig
ur
e
6
.
B
iL
S
T
M
a
r
c
hi
te
c
tu
r
e
2.7
.
O
p
t
im
iz
e
d
S
B
O
A
al
gor
it
h
m
S
e
c
r
e
ta
r
y
bi
r
d
is
a
n
A
f
r
ic
a
n
s
tr
ik
in
g
bi
r
d
w
hi
c
h
lo
oks
s
im
il
a
r
to
th
e
e
a
gl
e
bi
r
d
,
ha
vi
ng
w
in
g
s
of
gr
e
y
-
br
ow
n
f
e
a
th
e
r
s
,
c
he
s
t
is
w
hi
te
in
c
ol
or
,
a
nd
b
e
li
e
pa
r
t
is
de
e
p
bl
a
c
k
in
c
ol
or
.
T
h
e
m
os
t
p
e
c
ul
ia
r
c
ha
r
a
c
te
r
is
ti
c
of
s
e
c
r
e
ta
r
y bi
r
d i
s
i
ts
hunti
ng
s
ty
le
a
nd e
va
di
ng s
ty
le
w
hi
le
i
t
t
r
a
ve
ls
t
hr
ough gr
a
s
s
la
nds
.
T
he
s
e
two
na
tu
r
e
s
a
r
e
m
a
th
e
m
a
ti
c
a
ll
y
f
or
m
ul
a
te
d
in
th
e
f
or
m
of
e
xpl
or
a
ti
on
a
nd
e
xpl
oi
ta
ti
on
.
W
hi
le
it
pr
a
ys
th
e
s
na
ke
s
in
th
e
gr
a
s
s
la
nd
s
,
it
s
ha
unt
in
g
s
ty
le
is
d
e
f
in
e
d
a
s
e
xpl
or
a
ti
on
pha
s
e
in
th
e
gl
oba
l
s
e
a
r
c
h
s
pa
c
e
,
a
nd
w
hi
le
it
e
s
c
a
pe
s
f
r
om
th
e
r
ode
nt
s
,
it
i
s
c
a
ll
e
d
a
s
e
xpl
oi
ta
ti
on
in
t
he
lo
c
a
l
s
e
a
r
c
h
s
pa
c
e
.
B
ot
h
of
th
e
e
xpl
or
a
ti
on
a
nd
e
xpl
oi
ta
ti
on
a
r
e
th
e
m
a
jo
r
c
ont
r
ib
ut
io
ns
of
th
e
s
e
c
r
e
ta
r
y
bi
r
d
'
s
c
ha
r
a
c
te
r
is
ti
c
s
to
s
ol
v
in
g
th
e
opt
im
iz
a
ti
on
pr
obl
e
m
s
in
th
e
s
e
a
r
c
h
s
pa
c
e
.
O
ur
pr
opos
e
d
m
ode
l
u
s
e
s
th
i
s
opt
im
iz
e
d
a
lg
or
it
hm
to
r
e
duc
e
th
e
te
s
ti
ng
c
om
pl
e
xi
ty
a
nd t
o i
m
pr
ove
t
he
a
c
c
ur
a
c
y r
a
te
,
a
s
s
how
n i
n A
lg
o
r
it
hm
1
.
A
lg
or
it
m
1. C
N
N
-
B
iL
S
T
M
-
S
B
O
A
m
ode
l
I
nput
:
th
e
r
a
w
E
E
G
s
ig
na
l
O
ut
put
:
e
xt
r
a
c
te
d
s
e
iz
ur
e
E
E
G
s
ig
na
l
a
nd nor
m
a
l
E
E
G
s
ig
na
l
L
is
t
N
/
/
T
he
numbe
r
of
f
e
a
tu
r
e
s
=
1
∶
{
E
xt
r
a
c
t
th
e
f
e
a
tu
r
e
s
w
it
h P
C
A
S
to
r
e
t
he
t
im
e
s
e
r
ie
s
s
e
que
n
c
e
of
da
ta
us
in
g
B
iL
S
T
M
C
la
s
s
if
y t
he
f
e
a
tu
r
e
s
w
it
h C
N
N
O
pt
im
iz
e
t
he
pe
r
f
or
m
a
nc
e
us
in
g S
B
O
A
}
F
in
d t
he
r
e
s
ul
t
2.7.1.
M
at
h
e
m
at
ic
al
m
od
e
ll
in
g
T
he
m
a
th
e
m
a
ti
c
a
l
m
ode
ll
in
g of
t
he
S
B
O
A
i
s
pr
e
s
e
nt
e
d
a
s
f
ol
lo
w
s
:
i)
I
ni
ti
a
l
pha
s
e
:
t
he
r
a
ndom i
ni
ti
a
li
z
a
ti
on of
t
he
pos
it
io
n of
s
e
c
r
e
ta
r
y bi
r
d
is
in
(
1)
.
,
=
+
×
(
−
)
,
=
1
,
2
,
.
.
.
,
,
=
1
,
2
,
.
.
.
,
(
1)
W
he
r
e
is
th
e
pos
it
io
n
of
th
e
ℎ
S
e
c
r
e
ta
r
y
bi
r
d
a
nd
a
r
e
th
e
uppe
r
a
nd
lo
w
e
r
bounds
,
a
nd
r
is
a
r
a
ndom num
be
r
be
twe
e
n 0 a
nd 1
.
ii)
H
a
unt
in
g
s
tr
a
te
gy of
s
e
c
r
e
ta
r
y
bi
r
d (
e
xpl
o
r
a
ti
on pha
s
e
)
:
t
hi
s
ph
a
s
e
i
s
di
vi
de
d i
nt
o t
hr
e
e
s
ta
ge
s
:
s
e
a
r
c
hi
ng
th
e
pr
e
y,
in
ge
s
ti
ng
th
e
pr
e
y
,
a
nd
a
tt
a
c
ki
ng
th
e
pr
e
y
.
T
he
to
ta
l
ha
unt
in
g
ti
m
e
is
e
qua
ll
y
di
vi
de
d
in
to
ti
m
e
in
te
r
va
ls
a
s
i
n (
2)
.
<
1
3
,
1
3
<
<
2
3
a
nd
2
3
<
<
.
(
2)
W
he
r
e
t
is
t
he
c
ur
r
e
nt
i
te
r
a
ti
on
,
a
nd
T
is
th
e
to
ta
l
num
be
r
i
f
i
te
r
a
ti
on
.
iii)
U
pda
ti
ng s
e
c
r
e
ta
r
y bi
r
d’
s
pos
it
io
n
<
1
3
,
,
1
=
,
+
(
_
1
−
_
2
)
×
1
(
3)
=
{
,
1
,
1
<
,
,
C
ons
um
in
g t
he
pr
e
y i
s
m
a
th
e
m
a
ti
c
a
ll
y e
xpr
e
s
s
e
d a
s
(
4)
-
(
10)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8814
I
nt
J
A
dv A
ppl
S
c
i
,
V
ol
.
15, No
.
1, M
a
r
c
h 2026
:
198
-
208
204
=
(
1
,
)
(
4)
W
hi
le
1
3
<
<
2
3
,
,
1
=
+
(
(
⁄
)
⋀
4
)
×
(
−
0
.
5
)
×
(
−
,
)
(
5)
=
{
,
1
,
1
<
,
,
(
6)
W
hi
le
>
2
3
,
,
1
=
+
(
(
1
−
)
⋀
(
2
×
)
)
×
,
×
(
7)
=
{
,
1
,
1
<
,
,
(
8)
=
0
.
5
×
(
)
(
9)
(
)
=
×
×
|
|
1
(
10
)
W
he
r
e
S
is
a
f
ix
e
d c
ons
t
a
nt
.
T
he
n
is
de
f
in
e
d a
s
(
11)
.
=
[
Γ
(
1
+
)
×
(
2
)
Γ
(
1
+
2
)
×
×
2
(
−
1
2
)
]
1
(
11)
iv
)
E
s
c
a
pe
s
tr
a
te
gy
(
e
xpl
oi
ta
ti
on
pha
s
e
)
:
t
he
e
s
c
a
pe
s
tr
a
te
gy
in
tr
oduc
e
s
a
pe
r
tu
r
ba
ti
on
f
a
c
to
r
(
1
−
)
2
.
T
he
m
a
th
e
m
a
ti
c
a
l
f
or
m
ul
a
ti
on of
th
e
e
s
c
a
p
e
s
tr
a
te
gy i
s
(
12)
-
(
13)
.
,
2
=
{
1
:
+
(
2
×
−
1
)
×
(
1
−
)
2
×
,
,
<
2
:
,
,
+
2
×
(
−
×
,
,
)
,
(
12)
=
{
,
2
,
2
<
,
,
(
13)
3.
E
X
P
E
R
I
M
E
N
T
A
L
R
E
S
U
L
T
S
A
N
D
D
I
S
C
U
S
S
I
O
N
T
h
e
im
pl
e
m
e
nt
a
ti
on
a
nd
tr
a
i
ni
n
g
of
th
e
pr
o
po
s
e
d
C
N
N
-
B
iL
S
T
M
-
S
B
O
A
m
o
de
l
w
e
r
e
c
a
r
r
i
e
d
o
ut
o
n
a
hi
gh
-
p
e
r
f
or
m
a
nc
e
c
o
m
pu
ti
n
g
s
y
s
te
m
t
o
e
n
s
ur
e
e
f
f
ic
ie
nt
e
x
e
c
ut
i
on
a
n
d
a
c
c
u
r
a
t
e
c
onv
e
r
g
e
nc
e
.
T
h
e
s
y
s
t
e
m
w
a
s
e
qu
ip
pe
d w
it
h
a
n
I
nt
e
l
C
or
e
i
7
1
1t
h
g
e
n
e
r
a
ti
on
pr
oc
e
s
s
or
, 32
G
B
of
R
A
M
,
a
n
d a
n
N
V
I
D
I
A
G
e
F
or
c
e
R
T
X
3
080
G
P
U
w
it
h
10
G
B
of
de
di
c
a
t
e
d
m
e
m
or
y
f
or
a
c
c
e
l
e
r
a
t
e
d
c
o
m
put
a
t
io
n
.
T
h
e
m
od
e
l
w
a
s
d
e
v
e
l
op
e
d
u
s
in
g
P
yt
h
on 3.9
in
a
T
e
n
s
or
F
lo
w
a
n
d
K
e
r
a
s
e
nvi
r
on
m
e
nt
, e
xe
c
ut
e
d
o
n
U
b
un
tu
20.
04
L
T
S
.
A
ll
e
x
pe
r
im
e
nt
s
w
e
r
e
r
un
w
it
h
c
om
put
e
uni
f
ie
d
de
vi
c
e
a
r
c
h
it
e
c
t
ur
e
(
C
U
D
A
)
a
nd
C
U
D
A
d
e
e
p
ne
ur
a
l
ne
t
w
or
k
(
c
u
D
N
N
)
s
upp
or
t
to
le
v
e
r
a
g
e
G
P
U
-
ba
s
e
d
p
a
r
a
ll
e
l
pr
o
c
e
s
s
in
g
a
nd
r
e
du
c
e
t
r
a
i
ni
n
g t
im
e
.
T
o
tr
a
in
th
e
pr
opos
e
d
m
ode
l
e
f
f
e
c
ti
ve
ly
,
w
e
in
it
ia
ll
y
c
ol
le
c
te
d
a
to
ta
l
of
4,096
E
E
G
im
a
ge
s
.
T
o
e
nha
nc
e
th
e
d
a
ta
s
e
t'
s
di
ve
r
s
it
y
a
nd
im
pr
ove
m
ode
l
ge
n
e
r
a
li
z
a
t
io
n,
w
e
a
ppl
ie
d
s
ix
a
ugm
e
nt
a
ti
on
te
c
hni
que
s
,
r
e
s
ul
ti
ng
in
a
n
e
xp
a
nde
d
da
t
a
s
e
t
of
28,672
im
a
ge
s
.
T
hi
s
e
nr
i
c
he
d
da
ta
s
e
t
w
a
s
th
e
n
di
vi
de
d
in
to
tr
a
in
in
g,
te
s
ti
ng,
a
nd
va
li
da
ti
on
s
e
t
s
us
in
g
a
70:
15:
15
r
a
ti
o
.
T
hi
s
e
ns
ur
e
s
th
a
t
th
e
m
ode
l
is
tr
a
in
e
d
on
a
la
r
ge
por
ti
on
of
th
e
da
ta
w
hi
le
a
ls
o
b
e
in
g
e
va
lu
a
te
d
a
nd
v
a
li
da
te
d
on
s
e
pa
r
a
te
s
ubs
e
t
s
to
pr
e
ve
nt
ove
r
f
it
ti
ng
.
T
a
bl
e
5
il
lu
s
tr
a
te
s
th
e
da
ta
s
e
t
di
s
tr
ib
ut
io
n
a
c
r
os
s
e
a
c
h
c
a
te
gor
y
,
a
nd
T
a
bl
e
6
s
how
s
th
e
pe
r
f
or
m
a
nc
e
v
a
lu
e
s
of
th
e
vot
in
g m
ode
ls
w
it
h di
f
f
e
r
e
nt
t
e
s
t
dur
a
ti
ons
.
T
a
bl
e
5
.
S
pl
it
ti
ng w
it
h r
a
ti
o
S
e
t
P
e
r
c
e
nt
a
ge
(%)
N
um
be
r
of
i
m
a
ge
s
T
r
a
i
ni
ng
70
20,070
T
e
s
t
i
ng
15
4,301
V
a
l
i
da
t
i
on
15
4,301
T
ot
a
l
100
28,672
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
dv A
ppl
S
c
i
I
S
S
N
:
2252
-
8814
I
m
pr
ov
e
d s
e
i
z
ur
e
d
e
te
c
ti
on us
in
g opti
m
iz
e
d t
ime
s
e
qu
e
nc
e
ba
s
e
d de
e
p l
e
ar
ni
ng
…
(
P
us
panj
al
i
M
al
li
k
)
205
T
a
bl
e
6
.
A
c
c
ur
a
c
y
,
lo
s
s
,
a
nd va
l
-
lo
s
s
of
vot
in
g m
ode
ls
D
-
N
e
t
R
e
s
N
e
t
G
-
N
e
t
C
N
N
A
c
c
L
os
s
V
a
l
-
l
os
s
A
c
c
L
os
s
V
a
l
-
l
os
s
A
c
c
L
os
s
V
a
l
-
l
os
s
A
c
c
L
os
s
V
a
l
-
l
os
s
T
e
s
t
1
75.02
0.25
0.33
79.02
0.17
0.32
80.01
0.27
0.34
89.01
0.26
0.36
T
e
s
t
2
75.04
0.24
0.21
79.02
0.16
0.32
80.02
0.28
0.24
89.03
0.21
0.32
T
e
s
t
3
75.06
0.22
0.25
79.04
0.23
0.31
80.05
0.25
0.33
89.04
0.18
0.25
T
e
s
t
4
75.06
0.18
0.23
79.05
0.28
0.31
80.06
0.21
0.29
89.06
0.19
0.29
T
e
s
t
5
75.82
0.17
0.21
79.06
0.22
0.29
80.07
0.19
0.28
89.08
0.16
0.15
S
uc
c
e
s
s
f
ul
ly
r
un
th
e
m
ode
ls
li
s
te
d
in
T
a
bl
e
6
,
a
nd
a
f
te
r
c
o
m
pl
e
ti
on
,
it
is
f
ound
th
a
t
th
e
r
e
s
ul
t
obt
a
in
e
d
in
C
N
N
m
ode
l
s
upe
r
s
e
d
e
s
th
e
r
e
s
ul
t
s
w
it
h
D
-
N
e
t,
R
e
s
N
e
t
,
a
nd
G
-
N
e
t
.
T
he
a
c
c
ur
a
te
r
a
te
of
C
N
N
is
89.08,
lo
s
s
va
lu
e
is
0.16
,
a
nd
v
al
-
lo
s
s
v
a
lu
e
is
0.15
.
T
he
n
te
s
t
th
e
ot
he
r
pe
r
f
or
m
a
nc
e
m
e
tr
ic
s
to
pr
ove
th
e
s
upe
r
io
r
it
y
of
C
N
N
m
ode
l
a
m
ong
D
-
N
e
t,
R
e
s
N
e
t
,
a
nd
G
-
N
e
t
m
ode
ls
.
T
he
pe
r
f
or
m
a
nc
e
m
e
a
s
ur
e
s
in
te
r
m
s
of
a
c
c
ur
a
c
y, s
e
ns
it
iv
it
y, s
pe
c
if
ic
it
y
,
a
nd A
U
C
s
c
or
e
s
a
r
e
l
is
te
d i
n
T
a
bl
e
7.
T
a
bl
e
7
.
P
e
r
f
or
m
a
nc
e
m
e
a
s
ur
e
m
e
tr
ic
s
of
c
hos
e
n m
ode
l
s
M
ode
l
s
A
c
c
ur
a
c
y (
%
)
S
e
ns
i
t
i
vi
t
y (
%
)
S
pe
c
i
f
i
c
i
t
y (
%
)
M
C
C
A
U
C
s
c
or
e
D
-
N
e
t
75.82
78.62
76.09
78.03
0.77
R
e
s
N
e
t
79.06
79.19
79.28
80.32
0.79
G
-
N
e
t
80.07
80.09
81.03
80.04
0.80
C
N
N
89.08
89.01
90.01
89.09
0.89
A
f
te
r
a
na
ly
z
in
g
th
e
out
c
om
e
s
of
pe
r
f
or
m
a
nc
e
m
e
tr
ic
s
of
T
a
bl
e
s
6
a
nd
7,
w
e
f
ound
th
a
t
th
e
C
N
N
m
ode
l
is
th
e
be
s
t
c
ho
ic
e
of
vot
in
g
m
ode
l
f
or
E
E
G
s
ig
na
l
de
te
c
ti
on
w
it
h
th
e
B
onn
E
E
G
da
ta
s
e
t
.
A
lt
hough
th
e
r
e
s
ul
ts
of
T
a
bl
e
s
6
a
nd
7
a
r
e
s
uf
f
ic
ie
nt
to
pr
ove
th
a
t
C
N
N
c
a
n
be
pr
e
f
e
r
r
e
d
in
E
E
G
de
te
c
ti
on
,
a
nd
th
e
r
e
s
ul
ts
of
C
N
N
a
r
e
a
ls
o
m
a
t
c
hi
ng
w
it
h
th
e
r
e
s
ul
ts
of
th
e
e
xi
s
ti
n
g
m
ode
ls
,
it
la
c
ks
a
s
e
qu
e
nt
ia
l
ti
m
e
s
e
r
ie
s
r
e
pr
e
s
e
nt
a
ti
on
of
s
ig
na
ls
.
T
o
ove
r
c
om
e
th
is
,
w
e
in
te
gr
a
te
C
N
N
w
it
h
B
i
L
S
T
M
,
a
nd
th
e
hybr
id
C
N
N
-
B
iL
S
T
M
e
a
s
il
y pr
oc
e
s
s
t
he
l
ong s
e
qu
e
nc
e
of
t
im
e
s
e
r
ie
s
da
ta
.
T
h
e
a
c
c
ur
a
c
y r
a
te
of
C
N
N
s
ta
gna
te
d w
it
h 89.08%
,
w
hi
c
h
ne
e
ds
to
im
pr
ove
.
F
or
im
pr
ovi
ng
pe
r
f
or
m
a
nc
e
m
e
a
s
ur
e
s
of
hyb
r
id
C
N
N
-
Bi
L
S
T
M
m
ode
l,
w
e
c
om
bi
ne
it
w
it
h
th
e
opt
im
iz
a
ti
on a
lg
or
it
hm
, S
B
O
A
, a
nd t
he
out
c
om
e
s
of
pr
opos
e
d C
N
N
-
B
iL
S
T
M
-
S
B
O
A
i
s
l
i
s
te
d i
n
T
a
bl
e
8.
T
a
bl
e
8
.
P
r
opos
e
d C
N
N
-
B
iL
S
T
M
-
S
B
O
A
M
ode
l
A
c
c
ur
a
c
y (
%
)
S
e
ns
i
t
i
vi
t
y (
%
)
S
pe
c
i
f
i
c
i
t
y (
%
)
M
C
C
AUC
s
c
or
e
C
N
N
-
B
i
L
S
T
M
-
S
B
O
A
(
B
onn da
t
a
s
e
t
a
nd
A
-
E
c
l
a
s
s
)
98.49
96.05
97.03
97.01
0.97
T
o
e
va
lu
a
te
th
e
g
e
ne
r
a
li
z
a
ti
on
a
bi
li
ty
of
th
e
C
N
N
-
B
iL
S
T
M
-
S
B
O
A
m
ode
l,
k
-
f
ol
d
c
r
os
s
-
va
li
da
ti
on
w
a
s
e
m
pl
oye
d
.
T
he
d
a
ta
s
e
t
w
a
s
di
vi
de
d
in
to
k
s
ubs
e
t
s
,
a
nd
th
e
m
ode
l
w
a
s
tr
a
in
e
d
k
ti
m
e
s
,
e
a
c
h
ti
m
e
u
s
in
g
a
di
f
f
e
r
e
nt
s
ubs
e
t
a
s
th
e
va
li
da
ti
on
s
e
t
.
T
h
e
r
e
s
ul
t
is
pr
oduc
e
d
in
T
a
bl
e
9
.
T
hi
s
m
e
th
od
he
lp
s
to
e
ns
ur
e
r
obus
t
pe
r
f
or
m
a
nc
e
a
nd
pr
e
ve
nt
s
ove
r
f
it
ti
ng.
F
or
va
li
da
ti
ng
th
e
te
s
ti
ng
out
c
om
e
s
of
our
pr
opo
s
e
d
m
ode
l,
w
e
c
om
pa
r
e
th
e
pe
r
f
or
m
a
nc
e
m
e
a
s
ur
e
s
of
a
c
c
ur
a
c
y,
s
e
n
s
it
iv
it
y,
s
pe
c
if
ic
it
y,
a
nd
K
ul
lb
a
c
k
-
L
e
ib
le
r
(
K
L
)
di
ve
r
ge
nc
e
lo
s
s
f
unc
ti
on
va
lu
e
s
of
our
m
ode
l
w
it
h
th
e
e
xi
s
ti
ng
m
ode
ls
,
a
nd
th
e
r
e
s
ul
t
f
in
di
ngs
a
r
e
pr
e
s
e
nt
e
d
in
T
a
bl
e
10.
T
a
bl
e
9
.
9
-
f
ol
d va
li
da
ti
on ove
r
our
pr
opos
e
d m
ode
l
F
ol
d
A
c
c
ur
a
c
y (
%
)
S
e
ns
i
t
i
vi
t
y (
%
)
S
pe
c
i
f
i
c
i
t
y (
%
)
AUC
T
i
m
e
t
a
ke
n (
s
e
c
)
2
97.23
95.13
96.61
0.96
150
3
97.72
95.51
96.72
0.966
180
5
97.91
95.72
96.86
0.966
210
7
98.12
95.91
96.89
0.968
240
9
98.49
96.05
97.03
0.97
270
T
a
bl
e
10. P
e
r
f
or
m
a
nc
e
c
om
pa
r
is
on w
it
h e
xi
s
ti
ng mode
l
s
M
ode
l
s
A
c
c
ur
a
c
y
(%)
S
e
ns
i
t
i
vi
t
y
(%)
S
pe
c
i
f
i
c
i
t
y
(%)
A
U
C
s
c
or
e
K
L
d
i
ve
r
ge
nc
e
l
os
s
M
ul
t
i
s
c
a
l
e
c
onvol
ut
i
ona
l
[
6]
92.5
93.1
91.9
92.4
0.39
A
l
e
a
r
na
bl
e
a
nd e
xpl
a
i
na
bl
e
w
a
v
e
l
e
t
[
7]
89.3
88.4
88.7
0.88
0.48
D
e
e
p l
e
a
r
ni
ng
-
ba
s
e
d a
t
t
e
nt
i
on m
e
c
ha
ni
s
m
[
8]
98.38
-
-
-
0.33
C
ha
nne
l
-
w
e
i
ght
e
d s
pa
t
i
a
l
–
t
e
m
por
a
l
[
9]
97.23
-
-
-
0.32
P
r
opos
e
d C
N
N
-
B
i
L
S
T
M
-
S
B
O
A
(
A
-
E
c
l
a
s
s
)
98.49
96.05
97.03
0.97
0.29
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8814
I
nt
J
A
dv A
ppl
S
c
i
,
V
ol
.
15, No
.
1, M
a
r
c
h 2026
:
198
-
208
206
H
e
r
e
K
L
d
iv
e
r
ge
nc
e
l
os
s
f
unc
ti
on i
s
u
s
e
d t
o c
om
pa
r
e
t
w
o da
t
a
di
s
tr
ib
ut
io
ns
i
n
th
e
c
a
s
e
of
a
s
s
e
s
s
in
g
da
ta
s
e
t
a
nd mode
l
dr
if
t.
M
a
th
e
m
a
ti
c
a
ll
y, i
t
is
e
xpr
e
s
s
e
d a
s
(
14)
.
(
|
|
)
=
∑
(
)
×
l
o
g
(
(
)
(
)
)
(
14)
W
h
e
r
e
X
i
s
t
he
p
o
s
s
ib
l
e
e
v
e
nt
i
n
th
e
pr
ob
a
bi
li
t
y
s
e
a
r
c
h
s
pa
c
e
,
P
(
X
)
a
nd
Q
(
X
)
a
r
e
th
e
pr
ob
a
bi
li
t
ie
s
of
X
i
n
th
e
di
s
tr
i
but
io
n
of
P
a
nd
Q
, a
n
d t
h
e
r
a
t
io
(
)
(
)
⁄
r
e
pr
e
s
e
nt
s
th
e
li
ke
l
ih
ood
o
f
e
ve
nt
X
a
c
c
or
di
ng
to
P
c
om
p
a
r
e
d
to
Q
.
T
he
R
O
C
-
A
U
C
p
lo
t
a
nd
l
o
s
s
-
e
p
oc
h p
lo
t
of
t
h
e
pr
o
po
s
e
d
m
ode
l
a
r
e
d
e
p
ic
te
d i
n F
ig
ur
e
s
7
a
n
d 8
.
F
ig
ur
e
7
.
R
O
C
-
A
U
C
pl
ot
of
C
N
N
-
B
iL
S
T
M
-
S
B
O
A
m
ode
l
f
or
B
onn E
E
G
(
A
-
E
)
da
ta
s
e
t
F
ig
ur
e
8
.
M
ode
l
lo
s
s
of
C
N
N
-
B
iL
S
T
M
-
S
B
O
A
f
or
B
onn
(A
-
E
)
da
ta
s
e
t
4.
C
O
N
C
L
U
S
I
O
N
T
hi
s
s
tu
dy
im
pr
ove
s
th
e
E
E
G
de
te
c
ti
on
by
in
te
gr
a
ti
ng
de
e
p
le
a
r
ni
ng
w
it
h
a
n
opt
im
iz
a
ti
on
te
c
hni
que
.
T
he
a
ppl
ie
d
1D
s
e
que
nt
ia
l
C
N
N
pr
ove
s
it
s
s
upe
r
io
r
it
y
ove
r
ot
he
r
vot
in
g
m
ode
ls
w
it
h
th
e
te
s
ti
ng
a
c
c
ur
a
c
y
of
89.08%
,
s
e
ns
it
iv
it
y
of
89.01
%
,
s
pe
c
if
ic
it
y
o
f
90.01%
,
M
C
C
of
89.09%
,
a
nd
A
U
C
s
c
or
e
o
f
0.89
.
T
he
ti
m
e
s
e
que
nc
e
or
de
r
in
g
of
E
E
G
s
ig
na
ls
i
s
in
te
gr
a
te
d
w
it
h
a
B
iL
S
T
M
n
e
twor
k
to
e
nha
nc
e
th
e
s
e
que
nt
ia
l
da
ta
pr
oc
e
s
s
in
g
by
a
ddr
e
s
s
in
g
th
e
va
ni
s
hi
ng
gr
a
di
e
nt
pr
obl
e
m
.
F
ur
t
he
r
to
im
p
r
ove
th
e
te
s
t
a
c
c
ur
a
c
y,
to
r
e
duc
e
th
e
ove
r
-
f
it
ti
ng
pr
obl
e
m
,
a
nd
to
in
c
r
e
a
s
e
th
e
le
a
r
ni
ng
r
a
te
,
th
e
hybr
id
C
N
N
-
B
iL
S
T
M
m
ode
l
is
c
om
bi
ne
d
w
it
h
th
e
c
hos
e
n
S
B
O
A
opt
im
iz
a
ti
on
te
c
hni
que
,
yi
e
ld
in
g
a
n
a
c
c
ur
a
c
y
of
98.49%
,
s
e
n
s
it
iv
it
y
of
96.05%
,
s
pe
c
if
ic
it
y
of
97.03%
,
M
C
C
va
lu
e
of
97.01%
,
R
O
C
-
A
U
C
va
lu
e
of
0.97,
a
nd
K
L
di
ve
r
ge
nc
e
lo
s
s
va
lu
e
of
0.29
.
T
he
s
e
pe
r
f
or
m
a
nc
e
m
e
a
s
ur
e
s
of
th
e
pr
opos
e
d
C
N
N
-
B
iL
S
T
M
-
S
B
O
A
pr
ove
it
s
e
f
f
ic
ie
nc
y
in
e
pi
le
ps
y
de
te
c
ti
on
.
T
he
f
ut
ur
e
w
or
k w
il
l
f
oc
us
on i
m
pr
ove
d a
nom
a
ly
de
te
c
ti
on by s
e
pa
r
a
ti
ng nois
e
a
nd a
r
ti
f
a
c
ts
f
r
om
t
he
E
E
G
s
ig
na
ls
a
nd
in
te
gr
a
ti
ng
e
dge
de
vi
c
e
s
f
or
a
dva
nc
e
d c
li
ni
c
a
l
di
a
gno
s
is
.
F
U
N
D
I
N
G
I
N
F
O
R
M
A
T
I
O
N
T
hi
s
r
e
s
e
a
r
c
h
is
a
n
a
c
a
de
m
ic
w
or
k.
N
o
f
und
r
e
c
e
iv
e
d
e
it
he
r
f
r
om
in
te
r
na
l
or
e
xt
e
r
na
l
a
ge
nc
ie
s
to
c
a
r
r
y out t
hi
s
w
or
k
.
A
U
T
H
O
R
C
O
N
T
R
I
B
U
T
I
O
N
S
S
T
A
T
E
M
E
N
T
T
hi
s
jo
ur
na
l
us
e
s
th
e
C
ont
r
ib
ut
or
R
ol
e
s
T
a
xonomy
(
C
R
e
di
T
)
to
r
e
c
ogni
z
e
in
di
vi
dua
l
a
ut
hor
c
ont
r
ib
ut
io
ns
, r
e
duc
e
a
ut
hor
s
hi
p di
s
put
e
s
,
a
nd f
a
c
il
it
a
te
c
ol
la
bo
r
a
ti
on
.
N
am
e
o
f
A
u
t
h
or
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
P
us
pa
nj
a
li
M
a
ll
ik
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
A
ji
t
K
um
a
r
N
a
ya
k
✓
✓
✓
✓
✓
✓
S
a
ty
a
pr
a
ka
s
h S
w
a
in
✓
✓
✓
✓
✓
✓
✓
✓
✓
C
:
C
onc
e
pt
ua
l
i
z
a
t
i
on
M
:
M
e
t
hodol
ogy
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
da
t
i
on
Fo
:
Fo
r
m
a
l
a
na
l
ys
i
s
I
:
I
nve
s
t
i
ga
t
i
on
R
:
R
e
s
our
c
e
s
D
:
D
a
t
a
C
ur
a
t
i
on
O
:
W
r
i
t
i
ng
-
O
r
i
gi
na
l
D
r
a
f
t
E
:
W
r
i
t
i
ng
-
R
e
vi
e
w
&
E
di
t
i
ng
Vi
:
Vi
s
ua
l
i
z
a
t
i
on
Su
:
Su
pe
r
vi
s
i
on
P
:
P
r
oj
e
c
t
a
dm
i
ni
s
t
r
a
t
i
on
Fu
:
Fu
ndi
ng a
c
qui
s
i
t
i
on
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
dv A
ppl
S
c
i
I
S
S
N
:
2252
-
8814
I
m
pr
ov
e
d s
e
i
z
ur
e
d
e
te
c
ti
on us
in
g opti
m
iz
e
d t
ime
s
e
qu
e
nc
e
ba
s
e
d de
e
p l
e
ar
ni
ng
…
(
P
us
panj
al
i
M
al
li
k
)
207
C
O
N
F
L
I
C
T
O
F
I
N
T
E
R
E
S
T
S
T
A
T
E
M
E
N
T
A
ut
hor
s
s
ta
te
no c
onf
li
c
t
of
i
nt
e
r
e
s
t.
D
A
T
A
A
V
A
I
L
A
B
I
L
I
T
Y
T
he
da
ta
th
a
t
s
uppor
t
th
e
f
in
di
ngs
of
th
is
s
tu
dy
a
r
e
ope
nl
y
a
va
il
a
bl
e
in
th
e
B
onn
da
t
a
s
e
t
at
ht
tp
s
:/
/ww
w
.ukbonn.de
/e
pi
le
pt
ol
ogi
e
/a
r
be
it
s
gr
uppe
n/
a
g
-
le
hne
r
tz
-
ne
ur
ophys
ik
/d
ow
nl
oa
ds
/
.
R
E
F
E
R
E
N
C
E
S
[
1]
J
.
X
u,
K
.
Y
a
n,
Z
.
D
e
ng,
J
.
X
.
L
i
u, J
.
W
a
ng,
a
nd
S
.
Y
a
n,
“
E
E
G
-
ba
s
e
d
e
pi
l
e
pt
i
c
s
e
i
z
ur
e
de
t
e
c
t
i
on
us
i
ng
de
e
p
l
e
a
r
ni
ng
t
e
c
hni
que
s
:
a
s
ur
ve
y,”
N
e
u
r
oc
om
put
i
ng
, vol
. 610, 2024, doi
:
10.1016/
j
.ne
uc
om
.2024.128644.
[
2]
V
.
P
.
B
a
l
a
m
,
“
S
ys
t
e
m
a
t
i
c
r
e
vi
e
w
o
f
s
i
ngl
e
-
c
ha
nne
l
E
E
G
-
ba
s
e
d
dr
ow
s
i
ne
s
s
de
t
e
c
t
i
on
m
e
t
hods
,”
I
E
E
E
T
r
ans
ac
t
i
ons
on
I
nt
e
l
l
i
ge
nt
T
r
ans
por
t
at
i
on Sy
s
t
e
m
s
, vol
. 25, no. 11, pp. 15210
–
15228, 2024, doi
:
10.1109/
T
I
T
S
.2024.344224.
[
3]
P
.
M
a
l
l
i
k,
A
.
K
.
N
a
y
a
k,
S
.
K
.
M
oh
a
pa
t
r
a
,
a
nd
K
.
P
.
S
w
a
i
n,
“
S
H
-
O
S
P
:
a
hybr
i
d
a
l
gor
i
t
hm
us
i
ng
s
pot
t
e
d
hye
na
opt
i
m
i
z
e
r
e
na
bl
e
d
w
i
t
h
opt
i
m
a
l
s
t
oc
ha
s
t
i
c
pr
oc
e
s
s
f
or
e
pi
l
e
pt
i
c
s
e
i
z
ur
e
de
t
e
c
t
i
on,”
S
N
C
om
put
e
r
Sc
i
e
nc
e
,
vol
.
5,
no.
8,
2024,
doi
:
10.1007/
s
42979
-
024
-
03488
-
8.
[
4]
I
.
A
hm
a
d
e
t
al
.
,
“
A
n
e
f
f
i
c
i
e
nt
f
e
a
t
ur
e
s
e
l
e
c
t
i
on
a
nd
e
xpl
a
i
na
bl
e
c
l
a
s
s
i
f
i
c
a
t
i
on
m
e
t
hod
f
or
E
E
G
-
ba
s
e
d
e
pi
l
e
pt
i
c
s
e
i
z
ur
e
de
t
e
c
t
i
on,”
J
our
nal
of
I
nf
or
m
at
i
on Se
c
ur
i
t
y
and A
ppl
i
c
at
i
ons
, vol
. 80, 2024, doi
:
10.1016/
j
.j
i
s
a
.2023.103654.
[
5]
N
.
K
ha
l
i
d
a
nd
M
.
S
.
E
hs
a
n,
“
C
r
i
t
i
c
a
l
a
na
l
ys
i
s
of
P
a
r
ki
ns
on’
s
di
s
e
a
s
e
de
t
e
c
t
i
on
us
i
ng
E
E
G
s
ub
-
ba
nds
a
nd
ga
t
e
d
r
e
c
ur
r
e
nt
uni
t
,”
E
ngi
ne
e
r
i
ng Sc
i
e
nc
e
and T
e
c
hnol
ogy
, an I
nt
e
r
nat
i
onal
J
ou
r
nal
, vol
. 59, 2024,
doi
:
10.1016/
j
.j
e
s
t
c
h.2024.101855.
[
6]
L
.
Q
i
u,
J
.
L
i
,
I
.
Z
hong,
W
.
F
e
ng,
C
.
Z
hou,
a
nd
J
.
P
a
n,
“
A
nove
l
E
E
G
-
ba
s
e
d
P
a
r
ki
ns
on’
s
di
s
e
a
s
e
d
e
t
e
c
t
i
on
m
ode
l
us
i
ng
m
ul
t
i
s
c
a
l
e
c
onvol
ut
i
ona
l
pr
ot
ot
ype
ne
t
w
or
ks
,”
I
E
E
E
T
r
ans
ac
t
i
ons
on
I
ns
t
r
um
e
nt
at
i
on
and
M
e
as
ur
e
m
e
nt
,
vol
.
73,
pp.
1
–
14,
2024,
doi
:
10.1109/
T
I
M
.2024.3351248.
[
7]
Y
.
Y
u,
Y
.
L
i
,
Y
.
Z
hou,
Y
.
W
a
ng,
a
nd
J
.
W
a
ng,
“
A
l
e
a
r
na
bl
e
a
nd
e
xpl
a
i
na
bl
e
w
a
ve
l
e
t
ne
ur
a
l
ne
t
w
or
k
f
o
r
E
E
G
a
r
t
i
f
a
c
t
s
de
t
e
c
t
i
on
a
nd
c
l
a
s
s
i
f
i
c
a
t
i
on,”
I
E
E
E
T
r
ans
ac
t
i
ons
on
N
e
u
r
al
Sy
s
t
e
m
s
and
R
e
habi
l
i
t
at
i
on
E
ngi
ne
e
r
i
ng
,
vol
.
32,
pp.
3358
–
3368,
2024
,
doi
:
10.1109/
T
N
S
R
E
.2024.3452315.
[
8]
C
.
D
i
vva
l
a
a
nd
M
.
M
i
s
hr
a
,
“
D
e
e
p
l
e
a
r
ni
ng
-
ba
s
e
d
a
t
t
e
nt
i
on
m
e
c
ha
ni
s
m
f
or
a
ut
om
a
t
i
c
dr
ow
s
i
ne
s
s
de
t
e
c
t
i
on
us
i
ng
E
E
G
s
i
gna
l
,”
I
E
E
E
Se
ns
or
s
L
e
t
t
e
r
s
, vol
. 8, 2024, doi
:
10.1109/
L
S
E
N
S
.2024.3363735.
[
9]
X
.
L
i
,
J
.
T
a
ng,
X
.
L
i
,
a
nd
Y
.
Y
a
ng,
“
C
W
S
T
R
-
N
e
t
:
a
c
ha
nne
l
-
w
e
i
ght
e
d
s
p
a
t
i
a
l
–
t
e
m
por
a
l
r
e
s
i
dua
l
ne
t
w
or
k
ba
s
e
d
on
nons
m
oot
h
nonne
ga
t
i
ve
m
a
t
r
i
x
f
a
c
t
or
i
z
a
t
i
on
f
or
f
a
t
i
gue
de
t
e
c
t
i
on
u
s
i
ng
E
E
G
s
i
gna
l
s
,
”
B
i
om
e
di
c
al
Si
gnal
P
r
oc
e
s
s
i
ng
and
C
ont
r
ol
,
vol
.
97
,
2024, doi
:
10.1016/
j
.bs
pc
.2024.106685.
[
10]
S
.
R
ukhs
a
r
a
nd
A
.
K
.
T
i
w
a
r
i
,
“
L
i
ght
w
e
i
ght
c
onvol
ut
i
on
t
r
a
ns
f
or
m
e
r
f
o
r
c
r
o
s
s
-
pa
t
i
e
nt
s
e
i
z
ur
e
de
t
e
c
t
i
on
i
n
m
ul
t
i
-
c
ha
nne
l
E
E
G
s
i
gna
l
s
,”
C
o
m
put
e
r
M
e
t
hods
and P
r
og
r
am
s
i
n B
i
om
e
di
c
i
ne
, vol
. 242, 2023, doi
:
10.1016/
j
.c
m
pb.2023.107856.
[
11]
N
.
F
.
A
l
i
,
N
.
A
l
ba
s
t
a
ki
,
A
.
N
.
I
.
B
e
l
ka
c
e
m
,
M
.
E
.
A
de
l
,
a
nd
M
.
A
t
e
f
,
“
A
l
ow
-
c
om
pl
e
xi
t
y
c
om
bi
ne
d
e
nc
ode
r
-
L
S
T
M
-
a
t
t
e
nt
i
on
ne
t
w
or
ks
f
or
E
E
G
-
ba
s
e
d
de
pr
e
s
s
i
on
de
t
e
c
t
i
on,”
I
E
E
E
A
c
c
e
s
s
,
vol
.
12,
pp.
129390
–
129403,
2024,
doi
:
10.1109/
A
C
C
E
S
S
.2024.3436895.
[
12]
M
.
S
.
Y
a
m
a
m
ot
o
e
t
al
.
,
“
M
ode
l
l
i
ng
c
om
pl
e
x
E
E
G
da
t
a
di
s
t
r
i
but
i
on
on
t
he
R
i
e
m
a
nni
a
n
m
a
ni
f
ol
d
t
ow
a
r
d
out
l
i
e
r
de
t
e
c
t
i
on
a
nd
m
ul
t
i
m
oda
l
c
l
a
s
s
i
f
i
c
a
t
i
on,”
I
E
E
E
T
r
ans
ac
t
i
ons
on
B
i
om
e
di
c
al
E
ngi
ne
e
r
i
ng
,
vol
.
71,
no.
2,
pp.
377
–
387,
2024,
doi
:
10.1109/
T
B
M
E
.2023.3295769.
[
13]
T
.
I
s
l
a
m
,
M
.
B
a
s
a
k,
R
.
I
s
l
a
m
,
a
nd
A
.
D
.
R
oy,
“
I
nve
s
t
i
ga
t
i
ng
popul
a
t
i
on
-
s
pe
c
i
f
i
c
e
pi
l
e
ps
y
de
t
e
c
t
i
on
f
r
om
noi
s
y
E
E
G
s
i
gna
l
s
us
i
ng
de
e
p
-
l
e
a
r
ni
ng m
ode
l
s
,”
H
e
l
i
y
on
, vol
. 9, no. 12, 2023, doi
:
10.1016/
j
.he
l
i
yon.2023.e
22208.
[
14]
C
.
Z
ha
o,
R
.
P
e
ng,
a
nd
D
.
W
u,
“
S
our
c
e
-
f
r
e
e
dom
a
i
n
a
da
pt
a
t
i
on
(
S
F
D
A
)
f
or
pr
i
va
c
y
-
pr
e
s
e
r
vi
ng
s
e
i
z
ur
e
s
ubt
ype
c
l
a
s
s
i
f
i
c
a
t
i
on,
”
I
E
E
E
T
r
ans
ac
t
i
ons
on
N
e
ur
al
Sy
s
t
e
m
s
and
R
e
habi
l
i
t
at
i
on
E
ngi
ne
e
r
i
ng
,
vol
.
31,
pp.
2315
–
2325,
2023,
doi
:
10.1109/
T
N
S
R
E
.2023.3274563.
[
15]
H
.
S
.
O
ha
l
a
nd
S
.
M
a
nt
r
i
,
“
E
xpl
or
i
ng
E
E
G
-
ba
s
e
d
bi
om
a
r
ke
r
s
f
or
i
m
pr
ove
d
e
a
r
l
y
A
l
z
he
i
m
e
r
’
s
di
s
e
a
s
e
d
e
t
e
c
t
i
on:
a
f
e
a
t
ur
e
-
ba
s
e
d
a
ppr
oa
c
h ut
i
l
i
z
i
ng m
a
c
hi
ne
l
e
a
r
ni
ng,”
M
e
as
ur
e
m
e
nt
:
Se
ns
or
s
, vol
. 36, 2024, do
i
:
10.1016/
j
.m
e
a
s
e
n.2024.101403.
[
16]
S
.
K
.
K
ha
r
e
a
nd
U
.
R
.
A
c
ha
r
ya
,
“
A
da
z
d
-
N
e
t
:
a
ut
om
a
t
e
d
a
da
pt
i
ve
a
nd
e
xpl
a
i
na
bl
e
A
l
z
he
i
m
e
r
’
s
di
s
e
a
s
e
de
t
e
c
t
i
on
s
ys
t
e
m
u
s
i
ng
E
E
G
s
i
gna
l
s
,”
K
now
l
e
dge
-
B
as
e
d Sy
s
t
e
m
s
, vol
. 278, 2023, doi
:
10.1016/
j
.knos
y
s
.2023.110858.
[
17]
M
.
M
.
I
s
l
a
m
,
M
.
R
.
I
s
l
a
m
,
a
nd
M
.
S
.
I
s
l
a
m
,
“
A
n
e
f
f
i
c
i
e
nt
hum
a
n
c
om
pu
t
e
r
i
nt
e
r
a
c
t
i
on
t
hr
ough
ha
nd
ge
s
t
ur
e
us
i
ng
de
e
p
c
onvol
ut
i
ona
l
ne
ur
a
l
ne
t
w
or
k,”
SN
C
om
put
e
r
S
c
i
e
nc
e
, vol
. 1, 2020
, doi
:
10.100
7/
s
42979
-
020
-
00223
-
x
.
[
18]
Q
.
W
a
ng
e
t
al
.
,
“
A
hybr
i
d S
V
M
a
nd
ke
r
ne
l
f
unc
t
i
on
-
ba
s
e
d
s
pa
r
s
e
r
e
pr
e
s
e
nt
a
t
i
on
c
l
a
s
s
i
f
i
c
a
t
i
on
f
or
a
ut
om
a
t
e
d
e
pi
l
e
ps
y
de
t
e
c
t
i
on
i
n
E
E
G
s
i
gna
l
s
,”
N
e
ur
o
c
om
put
i
ng
, vol
. 562, 2023, doi
:
10.1016/
j
.ne
uc
om
.2023.126874.
[
19]
T
.
W
u,
Y
.
F
a
n,
Y
.
Z
hong,
X
.
C
he
ng,
X
.
K
ong,
a
nd
I
.
C
he
n,
“
S
C
N
e
t
:
a
s
pa
t
i
a
l
f
e
a
t
ur
e
f
us
e
d
c
onvol
ut
i
ona
l
ne
t
w
or
k
f
or
m
ul
t
i
-
c
ha
nne
l
E
E
G
pa
t
hol
ogy de
t
e
c
t
i
on,”
B
i
om
e
di
c
al
Si
gnal
P
r
oc
e
s
s
i
ng and C
ont
r
ol
, vol
. 86, 2023, doi
:
10.1016/
j
.bs
pc
.2023.105059.
[
20]
Y
.
F
u,
D
.
L
i
u,
J
.
C
he
n,
a
nd
I
.
H
e
,
“
S
e
c
r
e
t
a
r
y
bi
r
d
opt
i
m
i
z
a
t
i
on
a
l
gor
i
t
hm
:
a
ne
w
m
e
t
a
he
ur
i
s
t
i
c
f
or
s
ol
vi
ng
gl
ob
a
l
opt
i
m
i
z
a
t
i
on
pr
obl
e
m
s
,”
A
r
t
i
f
i
c
i
al
I
nt
e
l
l
i
ge
nc
e
R
e
v
i
e
w
, vol
. 57, 2024, doi
:
10.1007/
s
10462
-
024
-
10729
-
y.
[
21]
R
.
G
.
A
ndr
z
e
j
a
k,
K
.
L
e
hn
e
r
t
z
,
F
.
M
or
m
a
nn,
C
.
R
i
e
ke
,
P
.
D
a
vi
d,
a
nd
C
.
E
.
E
l
ge
r
,
“
I
ndi
c
a
t
i
ons
of
nonl
i
ne
a
r
de
t
e
r
m
i
ni
s
t
i
c
a
nd
f
i
ni
t
e
-
di
m
e
ns
i
ona
l
s
t
r
uc
t
ur
e
s
i
n
t
i
m
e
s
e
r
i
e
s
of
br
a
i
n
e
l
e
c
t
r
i
c
a
l
a
c
t
i
vi
t
y:
de
pe
nde
nc
e
on
r
e
c
or
di
ng
r
e
gi
on
a
nd
br
a
i
n
s
t
a
t
e
,”
P
hy
s
i
c
al
R
e
v
i
e
w
E
, vol
. 64, no. 6, 2001, doi
:
10.1103/
P
hys
R
e
vE
.64.061907.
[
22]
A
.
T
.
M
oha
n
a
nd
D
.
V
G
a
i
t
onde
,
“
A
de
e
p
l
e
a
r
ni
ng
b
a
s
e
d
a
ppr
oa
c
h
t
o
r
e
duc
e
d
or
de
r
m
ode
l
l
i
ng
f
or
t
ur
bul
e
nt
f
l
ow
c
ont
r
ol
us
i
n
g
L
S
T
M
ne
ur
a
l
ne
t
w
or
ks
,”
2018
,
a
r
X
i
v
:
1804.09269
.
[
23]
M
.
S
he
n,
P
.
W
e
n,
B
.
S
ong,
a
nd
Y
.
L
i
,
“
A
n
E
E
G
ba
s
e
d
r
e
a
l
-
t
i
m
e
e
pi
l
e
ps
y
s
e
i
z
ur
e
de
t
e
c
t
i
on
a
ppr
oa
c
h
u
s
i
ng
di
s
c
r
e
t
e
w
a
ve
l
e
t
t
r
a
ns
f
or
m
a
nd
m
a
c
hi
ne
l
e
a
r
ni
ng
m
e
t
hod
s
,”
B
i
om
e
di
c
al
Si
gnal
P
r
oc
e
s
s
i
ng
and
C
ont
r
ol
,
vol
.
77,
2022
,
doi
:
10.1016/
j
.bs
pc
.2022.103820.
[
24]
L
.
S
.
V
i
dya
r
a
t
ne
a
nd
K
.
M
.
I
f
t
e
kha
r
uddi
n,
“
R
e
a
l
-
t
i
m
e
e
pi
l
e
pt
i
c
s
e
i
z
ur
e
de
t
e
c
t
i
on
us
i
ng
E
E
G
,”
I
E
E
E
T
r
ans
ac
t
i
ons
on
N
e
ur
a
l
Sy
s
t
e
m
s
and R
e
habi
l
i
t
at
i
on E
ngi
ne
e
r
i
ng
, vol
. 25, pp. 2146
–
2156, 2017, doi
:
10.1109/
T
N
S
R
E
.2017.2697920.
[
25]
M
.
A
l
hus
s
e
i
n,
G
.
M
uha
m
m
a
d,
a
nd
M
.
S
.
H
o
s
s
a
i
n,
“
E
E
G
pa
t
hol
ogy
de
t
e
c
t
i
on
ba
s
e
d
on
de
e
p
l
e
a
r
ni
ng,”
I
E
E
E
A
c
c
e
s
s
,
vol
.
7,
pp. 27781
–
27788, 2019, doi
:
10.1109/
A
C
C
E
S
S
.2019.2901672.
[
26]
A
.
B
r
i
t
t
o
e
t
al
.
,
“
A
m
ul
t
i
-
di
m
e
ns
i
ona
l
hybr
i
d
C
N
N
-
B
i
L
S
T
M
f
r
a
m
e
w
or
k
f
or
e
pi
l
e
pt
i
c
s
e
i
z
ur
e
de
t
e
c
t
i
on
us
i
ng
e
l
e
c
t
r
oe
nc
e
pha
l
ogr
a
m
s
i
gna
l
s
c
r
ut
i
ny,”
Sy
s
t
e
m
s
and Sof
t
C
om
put
i
ng
, vol
. 5, 2023, doi
:
10.1016/
j
.s
a
s
c
.2023.200062.
Evaluation Warning : The document was created with Spire.PDF for Python.