I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
pu
t
er
E
ng
ineering
(
I
J
E
CE
)
Vo
l.
1
6
,
No
.
3
,
J
u
n
e
20
2
6
,
p
p
.
1562
~
1
5
7
5
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
1
6
i
3
.
pp
1
5
6
2
-
1
5
7
5
1562
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Perf
o
rma
nce
ana
ly
sis
of sing
le and
multi
-
sta
g
e m
eta
heuristic
o
ptimiza
tion o
n D
FF
NN
f
o
r
elect
ro
ca
rdio
g
ra
m
-
ba
sed emo
tion
cla
ss
ificatio
n
G
io
v
a
nn
i D
im
a
s
P
re
na
t
a
,
Ahm
a
d Ridho
’
i
F
a
c
u
l
t
y
o
f
I
n
t
e
l
l
i
g
e
n
t
El
e
c
t
r
i
c
a
l
a
n
d
I
n
f
o
r
mat
i
c
s
Te
c
h
n
o
l
o
g
y
,
U
n
i
v
e
r
s
i
t
a
s
1
7
A
g
u
st
u
s
1
9
4
5
S
u
r
a
b
a
y
a
,
S
u
r
a
b
a
y
a
,
I
n
d
o
n
e
s
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
an
1
6
,
2
0
2
6
R
ev
is
ed
Ma
r
2
7
,
2
0
2
6
Acc
ep
ted
Ap
r
2
6
,
2
0
2
6
Emo
ti
o
n
c
las
sifica
ti
o
n
b
a
se
d
o
n
e
lec
tro
c
a
rd
io
g
ra
m
(ECG
)
sig
n
a
ls
h
a
s
a
tt
ra
c
ted
in
c
re
a
sin
g
a
tt
e
n
ti
o
n
i
n
a
ffe
c
ti
v
e
c
o
m
p
u
ti
n
g
a
n
d
b
i
o
m
e
d
i
c
a
l
sig
n
a
l
p
ro
c
e
ss
in
g
.
H
o
we
v
e
r,
train
i
n
g
d
e
e
p
fe
e
d
fo
rwa
rd
n
e
u
ra
l
n
e
two
rk
s
(DFF
NN
)
u
sin
g
c
o
n
v
e
n
ti
o
n
a
l
g
ra
d
ien
t
-
b
a
se
d
lea
rn
in
g
o
ften
su
ffe
rs
fr
o
m
lo
c
a
l
m
in
ima
a
n
d
sl
o
w
c
o
n
v
e
r
g
e
n
c
e
,
p
a
rti
c
u
larl
y
wh
e
n
d
e
a
li
n
g
wit
h
n
o
n
li
n
e
a
r
a
n
d
li
m
it
e
d
d
a
tas
e
ts.
Th
is
stu
d
y
p
re
se
n
ts
a
c
o
m
p
re
h
e
n
si
v
e
p
e
rfo
rm
a
n
c
e
a
n
a
ly
sis
o
f
sin
g
le
-
sta
g
e
a
n
d
m
u
l
ti
-
sta
g
e
m
e
tah
e
u
risti
c
o
p
ti
m
iza
ti
o
n
stra
teg
ies
a
p
p
li
e
d
t
o
DFF
NN
fo
r
ECG
-
b
a
se
d
e
m
o
ti
o
n
c
las
sifica
ti
o
n
in
e
ld
e
rly
p
a
rti
c
i
p
a
n
ts.
F
iv
e
m
o
d
e
ls
we
re
e
v
a
lu
a
ted
:
P
u
re
D
F
F
NN
,
DFF
NN
o
p
t
imiz
e
d
u
si
n
g
g
e
n
e
ti
c
a
lg
o
rit
h
m
(G
A),
p
a
rti
c
le
sw
a
rm
o
p
t
imiz
a
ti
o
n
(
P
S
O),
g
re
y
w
o
lf
o
p
t
imiz
e
r
(G
WO),
a
n
d
a
h
y
b
rid
m
u
lt
i
-
sta
g
e
DFF
NN
+
G
A+
G
WO
m
o
d
e
l.
Ex
p
e
rime
n
tal
re
su
lt
s
fro
m
six
in
d
e
p
e
n
d
e
n
t
tri
a
ls
d
e
m
o
n
stra
te
a
su
b
sta
n
ti
a
l
re
d
u
c
ti
o
n
i
n
m
e
a
n
sq
u
a
re
d
e
rro
r
(M
S
E)
wh
e
n
m
e
tah
e
u
risti
c
o
p
ti
m
iza
ti
o
n
is
a
p
p
li
e
d
.
P
u
re
DFF
NN
p
ro
d
u
c
e
d
fin
a
l
M
S
E
v
a
lu
e
s
in
th
e
ra
n
g
e
o
f
0
.
0
7
4
6
2
–
0
.
0
8
9
7
7
,
wh
e
re
a
s
DFF
NN
+
G
WO
re
d
u
c
e
d
M
S
E
to
0
.
0
1
8
9
4
–
0
.
0
2
4
1
1
.
Th
e
p
ro
p
o
se
d
m
u
lt
i
-
sta
g
e
DFF
NN
+
G
A+
G
W
O
a
c
h
iev
e
d
th
e
lo
we
st
M
S
E
o
f
0
.
0
1
4
2
8
6
i
n
th
e
b
e
st
ru
n
a
n
d
a
n
a
v
e
ra
g
e
M
S
E
o
f
a
p
p
ro
x
ima
tely
0
.
0
2
1
2
a
c
ro
ss
tri
a
ls
.
Train
in
g
a
c
c
u
ra
c
y
imp
ro
v
e
d
fr
o
m
5
7
.
1
4
%
–
6
6
.
6
7
%
(P
u
re
DFF
NN
)
to
8
0
.
9
5
%
–
8
5
.
7
1
%
u
sin
g
m
e
tah
e
u
ri
stic
a
p
p
ro
a
c
h
e
s.
Alt
h
o
u
g
h
tes
ti
n
g
a
c
c
u
ra
c
y
re
m
a
in
e
d
re
lativ
e
ly
sta
b
le
a
t
3
3
.
3
3
%
–
5
0
.
0
0
%
d
u
e
t
o
d
a
tas
e
t
siz
e
c
o
n
stra
in
ts,
c
o
n
v
e
rg
e
n
c
e
b
e
h
a
v
i
o
r
a
n
a
ly
sis
sh
o
ws
th
a
t
m
u
lt
i
-
sta
g
e
o
p
ti
m
iza
ti
o
n
e
n
h
a
n
c
e
s
sta
b
il
it
y
a
n
d
re
d
u
c
e
s
o
sc
il
lat
o
ry
u
p
d
a
tes
.
Th
e
se
fin
d
in
g
s
c
o
n
firm
th
a
t
m
u
lt
i
-
sta
g
e
m
e
tah
e
u
risti
c
o
p
ti
m
iza
ti
o
n
sig
n
ifi
c
a
n
t
ly
imp
r
o
v
e
s
train
i
n
g
st
a
b
il
it
y
a
n
d
e
rro
r
m
in
imiz
a
ti
o
n
i
n
DFF
NN
m
o
d
e
ls,
o
ffe
rin
g
a
p
ro
m
isin
g
st
ra
teg
y
fo
r
ro
b
u
st E
CG
-
b
a
se
d
e
m
o
ti
o
n
c
las
sifica
ti
o
n
u
n
d
e
r
sm
a
ll
-
sa
m
p
le co
n
d
i
ti
o
n
s.
K
ey
w
o
r
d
s
:
Dee
p
f
ee
d
f
o
r
war
d
n
eu
r
al
n
etwo
r
k
E
lectr
o
ca
r
d
io
g
r
am
E
m
o
tio
n
class
if
icatio
n
Gen
etic
alg
o
r
ith
m
Gr
ey
wo
lf
o
p
tim
izer
Hy
b
r
id
m
etah
e
u
r
is
tics
Par
ticle
s
war
m
o
p
tim
izatio
n
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
:
Gio
v
an
n
i D
im
as Pr
en
ata
Facu
lty
o
f
I
n
tellig
en
t E
lectr
ical
an
d
I
n
f
o
r
m
atics T
ec
h
n
o
lo
g
y
,
Un
iv
er
s
i
tas 1
7
Ag
u
s
tu
s
1
9
4
5
Su
r
ab
ay
a
Su
r
ab
ay
a,
I
n
d
o
n
esia
E
m
ail: g
p
r
en
ata@u
n
tag
-
s
b
y
.
a
c.
id
1.
I
NT
RO
D
UCT
I
O
N
E
lectr
o
ca
r
d
io
g
r
am
(
E
C
G)
s
ig
n
als
d
o
n
o
t
m
er
ely
r
ep
r
esen
t
t
h
e
elec
tr
ical
ac
tiv
ity
o
f
t
h
e
h
e
ar
t,
b
u
t
also
en
ca
p
s
u
late
r
ich
p
h
y
s
io
lo
g
ical
in
f
o
r
m
atio
n
th
at
is
cl
o
s
ely
as
s
o
ciate
d
with
h
u
m
an
em
o
tio
n
al
s
tates.
Nu
m
er
o
u
s
s
tu
d
ies
h
av
e
d
em
o
n
s
tr
ated
th
a
t
em
o
tio
n
al
v
ar
iatio
n
s
s
u
ch
as
s
tr
ess
,
an
x
iety
,
f
ea
r
,
an
d
r
elax
atio
n
in
f
lu
en
ce
th
e
d
y
n
am
ics
o
f
th
e
au
to
n
o
m
ic
n
e
r
v
o
u
s
s
y
s
tem
,
w
h
ich
a
r
e
r
e
f
lecte
d
in
c
h
an
g
es
i
n
E
C
G
m
o
r
p
h
o
lo
g
y
an
d
r
h
y
th
m
,
p
ar
ticu
lar
ly
th
r
o
u
g
h
h
ea
r
t
r
at
e
v
ar
iab
ilit
y
(
HR
V)
p
ar
a
m
eter
s
[
1
]
,
[
2
]
.
C
o
n
s
eq
u
en
tly
,
E
C
G
-
b
ased
em
o
tio
n
class
if
icatio
n
h
as
em
er
g
ed
a
s
an
in
cr
ea
s
in
g
ly
ac
tiv
e
r
ese
ar
ch
to
p
ic,
esp
ec
ially
in
a
p
p
licatio
n
s
r
elate
d
to
af
f
ec
tiv
e
co
m
p
u
tin
g
,
m
en
tal
h
ea
lth
m
o
n
ito
r
in
g
,
an
d
h
u
m
an
c
o
m
p
u
ter
i
n
ter
ac
tio
n
[
3
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
P
erfo
r
ma
n
ce
a
n
a
lysi
s
o
f sin
g
le
a
n
d
mu
lti
-
s
ta
g
e
meta
h
eu
r
is
tic
o
p
timiz
a
tio
n
o
n
…
(
Gio
va
n
n
i D
ima
s
P
r
en
a
ta
)
1563
R
ec
en
t
ad
v
an
ce
s
in
elec
tr
o
c
ar
d
io
g
r
a
m
an
aly
s
is
h
av
e
also
ex
p
an
d
ed
f
r
o
m
c
o
n
v
e
n
tio
n
al
em
o
tio
n
r
ec
o
g
n
itio
n
to
war
d
s
ig
n
al
f
o
r
ec
asti
n
g
,
d
is
ea
s
e
d
etec
tio
n
,
a
n
d
d
ee
p
s
eq
u
en
tial
lear
n
in
g
f
r
a
m
ewo
r
k
s
.
Z
ac
ar
ias
et
a
l.
[
4
]
d
em
o
n
s
tr
ated
t
h
at
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
L
S
T
M)
n
etwo
r
k
s
ca
n
ef
f
ec
tiv
el
y
m
o
d
el
tem
p
o
r
al
d
ep
en
d
e
n
cies
in
E
C
G
f
o
r
ec
asti
n
g
task
s
.
Hy
b
r
id
p
r
e
d
ictio
n
ap
p
r
o
ac
h
es
co
m
b
in
in
g
s
ig
n
al
d
ec
o
m
p
o
s
itio
n
an
d
n
eu
r
al
n
etwo
r
k
s
h
av
e
also
s
h
o
wn
p
r
o
m
is
in
g
p
er
f
o
r
m
an
ce
f
o
r
E
C
G
r
ec
o
n
s
tr
u
ctio
n
an
d
p
r
e
d
ictio
n
[
5
]
.
Fu
r
th
er
m
o
r
e
,
r
ec
en
t
s
y
s
tem
atic
r
ev
iews
co
n
f
ir
m
th
at
d
ee
p
lear
n
in
g
m
et
h
o
d
s
ar
e
in
cr
ea
s
in
g
ly
d
o
m
in
a
n
t
in
E
C
G
clas
s
if
icatio
n
d
u
e
to
th
eir
ab
ilit
y
to
au
to
m
atica
lly
lear
n
co
m
p
le
x
r
ep
r
esen
tatio
n
s
f
r
o
m
p
h
y
s
io
lo
g
ical
d
ata
[
6
]
.
C
o
n
v
en
tio
n
al
a
p
p
r
o
ac
h
es
r
el
y
in
g
o
n
h
an
d
c
r
af
ted
s
tatis
tical
f
ea
tu
r
es
an
d
lin
ea
r
class
if
icatio
n
m
eth
o
d
s
o
f
ten
f
ac
e
lim
itatio
n
s
in
ca
p
tu
r
in
g
th
e
n
o
n
lin
ea
r
an
d
co
m
p
lex
ch
ar
ac
ter
is
tics
o
f
E
C
G
s
ig
n
als.
T
o
ad
d
r
ess
th
is
is
s
u
e,
ar
tific
ial
n
eu
r
al
n
etwo
r
k
(
ANN)
-
b
ased
m
o
d
els
h
av
e
b
ee
n
wid
ely
ad
o
p
te
d
d
u
e
to
t
h
e
ir
s
tr
o
n
g
ca
p
ab
ilit
y
in
m
o
d
elin
g
n
o
n
lin
ea
r
r
elati
o
n
s
h
ip
s
am
o
n
g
p
h
y
s
io
lo
g
ical
s
ig
n
al
f
ea
tu
r
es
[
7
]
.
On
e
r
elativ
ely
s
im
p
le
y
et
ef
f
ec
tiv
e
ANN
ar
ch
itectu
r
e
i
s
th
e
d
ee
p
f
ee
d
f
o
r
war
d
n
eu
r
al
n
etwo
r
k
(
DFFNN)
,
wh
ich
em
p
lo
y
s
m
u
ltip
le
h
id
d
en
lay
er
s
to
e
n
h
an
c
e
r
ep
r
esen
tatio
n
al
ca
p
ac
ity
[
8
]
.
Ne
v
er
th
eless
,
th
e
p
er
f
o
r
m
an
ce
o
f
DFFNN
is
h
ig
h
ly
d
ep
en
d
e
n
t o
n
th
e
tr
ain
in
g
p
r
o
c
ess
,
p
ar
ticu
lar
ly
o
n
t
h
e
o
p
tim
i
za
tio
n
o
f
n
etwo
r
k
weig
h
ts
an
d
b
iases
.
Sev
er
al
r
ec
en
t
s
tu
d
ies
h
a
v
e
also
ap
p
lied
r
ec
u
r
r
en
t
n
eu
r
al
n
etwo
r
k
s
an
d
L
STM
a
r
ch
it
ec
tu
r
es
f
o
r
E
C
G
-
b
ased
d
is
ea
s
e
r
ec
o
g
n
itio
n
an
d
b
io
m
ed
ical
class
if
icatio
n
,
s
h
o
win
g
s
tr
o
n
g
ca
p
ab
ilit
y
i
n
tem
p
o
r
al
f
ea
tu
r
e
m
o
d
elin
g
[
9
]
.
Ho
wev
er
,
th
es
e
m
eth
o
d
s
o
f
ten
r
eq
u
ir
e
lar
g
er
d
atasets
an
d
h
ig
h
er
co
m
p
u
tatio
n
al
co
m
p
lex
ity
th
an
f
ee
d
f
o
r
war
d
ar
ch
itectu
r
es
,
m
ak
in
g
DFFNN
s
till
a
ttra
cti
v
e
f
o
r
s
m
all
-
s
am
p
le
s
ce
n
ar
io
s
s
u
ch
as
th
e
p
r
esen
t
s
tu
d
y
.
Gr
ad
ien
t
-
b
ased
lear
n
in
g
m
eth
o
d
s
s
u
ch
as
b
ac
k
p
r
o
p
ag
atio
n
f
r
eq
u
e
n
tly
s
u
f
f
er
f
r
o
m
is
s
u
es
in
clu
d
in
g
lo
ca
l
m
in
im
a
e
n
tr
ap
m
e
n
t,
s
lo
w
co
n
v
e
r
g
en
ce
,
an
d
s
en
s
itiv
ity
to
weig
h
t
in
itializatio
n
,
esp
e
cially
wh
en
d
ea
lin
g
with
n
o
is
y
an
d
n
o
n
-
s
tatio
n
ar
y
E
C
G
d
ata
[
1
0
]
.
T
h
ese
lim
itatio
n
s
h
av
e
m
o
tiv
ated
th
e
ad
o
p
t
io
n
o
f
p
o
p
u
latio
n
-
b
ased
m
etah
eu
r
is
tic
o
p
tim
izatio
n
alg
o
r
ith
m
s
th
at
d
o
n
o
t
r
ely
o
n
g
r
a
d
ien
t
in
f
o
r
m
atio
n
.
Am
o
n
g
th
e
m
o
s
t
wid
ely
ap
p
lied
m
etah
e
u
r
is
tic
tech
n
iq
u
es
f
o
r
n
eu
r
al
n
etwo
r
k
weig
h
t
o
p
tim
izatio
n
ar
e
th
e
g
en
etic
al
g
o
r
ith
m
(
GA)
,
p
ar
ticle
s
war
m
o
p
tim
iza
tio
n
(
PS
O)
,
an
d
g
r
ey
w
o
lf
o
p
ti
m
izer
(
GW
O)
[
1
1
]
–
[
1
3
]
.
GA
m
im
ics
b
io
lo
g
ical
ev
o
l
u
tio
n
ar
y
p
r
o
ce
s
s
es
s
u
ch
as
s
elec
tio
n
,
c
r
o
s
s
o
v
er
,
an
d
m
u
tatio
n
t
o
p
er
f
o
r
m
g
lo
b
al
s
ea
r
ch
i
n
co
m
p
lex
s
o
lu
tio
n
s
p
ac
es
[
1
4
]
.
W
h
ile
G
A
is
ef
f
ec
tiv
e
in
ex
p
lo
r
atio
n
,
it
o
f
ten
ex
h
ib
its
r
elativ
ely
s
lo
w
co
n
v
er
g
en
ce
d
u
r
in
g
t
h
e
ex
p
lo
itatio
n
p
h
ase.
I
n
co
n
tr
ast,
PSO
,
in
s
p
ir
ed
b
y
t
h
e
s
o
cial
b
eh
a
v
io
r
o
f
p
ar
ticle
s
war
m
s
,
is
k
n
o
wn
f
o
r
its
f
ast
co
n
v
er
g
en
ce
s
p
ee
d
,
b
u
t
it
is
s
u
s
ce
p
tib
le
t
o
p
r
em
atu
r
e
co
n
v
er
g
en
ce
[
1
5
]
.
GW
O
,
wh
ich
is
in
s
p
ir
ed
b
y
th
e
s
o
cial
h
ier
ar
ch
y
an
d
h
u
n
tin
g
s
tr
ateg
ies o
f
g
r
ey
w
o
lv
e
s
,
h
as d
em
o
n
s
tr
ated
a
f
av
o
r
ab
le
b
alan
ce
b
etwe
en
e
x
p
lo
r
atio
n
an
d
e
x
p
lo
itatio
n
,
al
o
n
g
with
s
tr
o
n
g
co
n
v
er
g
en
ce
s
tab
ilit
y
in
v
ar
io
u
s
n
o
n
lin
ea
r
o
p
tim
izatio
n
p
r
o
b
lem
s
[
1
6
]
.
I
n
th
e
co
n
tex
t
o
f
E
C
G
-
b
ased
em
o
tio
n
class
if
icatio
n
u
s
in
g
n
eu
r
al
n
etwo
r
k
s
,
m
o
s
t
ex
is
tin
g
s
tu
d
ies
h
av
e
p
r
im
ar
ily
f
o
cu
s
ed
o
n
s
in
g
le
-
s
tag
e
m
etah
eu
r
is
tic
o
p
tim
izatio
n
,
wh
er
e
o
n
ly
o
n
e
o
p
tim
i
za
tio
n
alg
o
r
ith
m
is
in
teg
r
ated
with
ar
tific
ial
n
eu
r
al
n
etwo
r
k
o
r
d
ee
p
n
eu
r
al
n
et
wo
r
k
s
(
DNN)
m
o
d
els
[
1
7
]
–
[
1
9
]
.
Alth
o
u
g
h
s
u
ch
ap
p
r
o
ac
h
es
g
en
er
ally
o
u
t
p
er
f
o
r
m
co
n
v
en
tio
n
al
g
r
ad
ien
t
-
b
a
s
ed
tr
ain
in
g
,
th
ey
m
ay
s
till
e
x
h
ib
it
co
n
v
er
g
e
n
ce
in
s
tab
ilit
y
an
d
s
ig
n
if
ican
t
f
lu
c
tu
atio
n
s
in
m
ea
n
s
q
u
a
r
ed
er
r
o
r
(
MSE
)
d
u
r
in
g
th
e
tr
ain
in
g
p
r
o
ce
s
s
,
p
ar
ticu
lar
ly
wh
en
ap
p
lied
to
co
m
p
lex
p
h
y
s
io
lo
g
ical
s
ig
n
al
d
atasets
.
As
an
alter
n
ativ
e,
m
u
lti
-
s
tag
e
m
etah
eu
r
is
tic
o
p
tim
izatio
n
h
a
s
r
ec
en
tly
attr
ac
ted
in
cr
ea
s
in
g
atten
tio
n
.
I
n
th
is
s
tr
ateg
y
,
o
n
e
alg
o
r
ith
m
is
em
p
lo
y
ed
d
u
r
in
g
th
e
in
i
tial
tr
ain
in
g
s
tag
e
to
en
h
an
ce
g
lo
b
al
ex
p
lo
r
atio
n
,
f
o
llo
wed
b
y
an
o
th
er
al
g
o
r
ith
m
th
at
em
p
h
asizes
lo
ca
l
ex
p
lo
itatio
n
in
later
s
tag
es
[
2
0
]
.
C
o
n
ce
p
tu
ally
,
t
h
is
ap
p
r
o
ac
h
aim
s
to
co
m
b
in
e
th
e
s
tr
en
g
th
s
o
f
d
if
f
er
e
n
t
o
p
tim
i
za
tio
n
alg
o
r
ith
m
s
,
lead
in
g
to
f
aster
co
n
v
er
g
e
n
ce
,
im
p
r
o
v
e
d
s
tab
ilit
y
,
an
d
l
o
wer
f
in
al
MSE
v
alu
es.
I
n
o
u
r
p
r
ev
io
u
s
wo
r
k
,
m
etah
e
u
r
is
tic
-
o
p
tim
ized
n
eu
r
al
n
etw
o
r
k
s
wer
e
s
u
cc
ess
f
u
lly
ap
p
lied
to
E
C
G
p
r
ed
ictio
n
u
n
d
er
lim
ited
-
d
ata
co
n
d
itio
n
s
,
wh
er
e
h
y
b
r
id
o
p
tim
izatio
n
s
tr
ateg
ies
s
h
o
wed
im
p
r
o
v
e
d
p
r
e
d
ictiv
e
s
tab
ilit
y
co
m
p
ar
ed
with
b
aselin
e
m
o
d
els
[
2
1
]
.
T
h
ese
f
in
d
in
g
s
m
o
tiv
ate
th
e
p
r
esen
t
ex
ten
s
io
n
to
war
d
E
C
G
-
b
ased
em
o
tio
n
class
if
icatio
n
.
Fro
m
T
ab
le
1
b
ased
o
n
th
e
l
im
itatio
n
s
o
f
ex
is
tin
g
s
tu
d
ies,
th
is
r
esear
ch
p
r
o
p
o
s
es
a
co
m
p
ar
ativ
e
f
r
am
ewo
r
k
in
v
o
l
v
in
g
s
in
g
le
-
s
tag
e
an
d
m
u
lti
-
s
tag
e
m
etah
eu
r
is
tic
o
p
tim
izatio
n
ap
p
lied
to
DFFNN
f
o
r
E
C
G
-
b
ased
em
o
tio
n
class
if
icatio
n
.
T
h
er
ef
o
r
e,
th
is
s
tu
d
y
s
h
o
u
l
d
b
e
in
te
r
p
r
eted
n
o
t
m
er
ely
as
a
class
if
icatio
n
p
er
f
o
r
m
an
ce
co
m
p
ar
is
o
n
,
b
u
t
as
an
an
aly
s
is
o
f
o
p
tim
izati
o
n
s
tab
ilit
y
an
d
c
o
n
v
e
r
g
en
ce
ch
ar
ac
ter
is
tics
in
m
etah
eu
r
is
tic
-
d
r
iv
en
n
eu
r
al
n
e
two
r
k
s
.
B
ey
o
n
d
s
in
g
le
o
p
tim
izer
s
,
r
ec
en
t
en
g
i
n
ee
r
in
g
s
tu
d
ies
h
av
e
s
h
o
wn
th
at
h
y
b
r
i
d
s
war
m
-
b
ased
m
etah
eu
r
is
tics
ca
n
im
p
r
o
v
e
e
x
p
lo
r
atio
n
-
ex
p
l
o
itatio
n
b
alan
c
e
m
o
r
e
ef
f
ec
tiv
ely
th
a
n
s
tan
d
alo
n
e
m
eth
o
d
s
.
Fo
r
ex
am
p
le,
c
o
m
b
in
atio
n
s
o
f
G
W
O
an
d
PS
O
h
av
e
d
em
o
n
s
tr
a
ted
im
p
r
o
v
ed
r
o
b
u
s
tn
ess
in
c
o
m
p
lex
o
p
tim
izatio
n
p
r
o
b
lem
s
[
2
2
]
.
T
h
is
s
u
p
p
o
r
ts
th
e
r
atio
n
ale
f
o
r
in
v
esti
g
atin
g
s
tag
ed
h
y
b
r
id
o
p
tim
izatio
n
s
tr
ateg
ies
in
n
eu
r
al
n
etwo
r
k
tr
ain
i
n
g
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
1
6
,
No
.
3
,
J
u
n
e
20
2
6
:
1
5
6
2
-
1
5
7
5
1564
T
ab
le
1
.
Su
m
m
a
r
y
o
f
r
elate
d
wo
r
k
o
n
E
C
G
-
B
ased
em
o
tio
n
class
if
icatio
n
Ref
M
e
t
h
o
d
D
a
t
a
s
e
t
O
p
t
i
mi
z
a
t
i
o
n
A
c
c
u
r
a
c
y
Li
mi
t
a
t
i
o
n
[
2
]
ANN
D
EA
P
N
o
n
e
7
8
%
Lo
c
a
l
mi
n
i
m
a
[
1
8
]
C
N
N
EC
G
d
a
t
a
se
t
G
r
a
d
i
e
n
t
8
5
%
La
r
g
e
d
a
t
a
r
e
q
u
i
r
e
me
n
t
[
1
9
]
S
V
M
H
R
V
f
e
a
t
u
r
e
s
G
r
i
d
s
e
a
r
c
h
8
0
%
Li
mi
t
e
d
s
c
a
l
a
b
i
l
i
t
y
[
1
6
]
M
LP+G
W
O
U
C
I
G
W
O
8
7
%
S
i
n
g
l
e
-
st
a
g
e
o
n
l
y
P
u
r
p
o
se
D
F
F
N
N
+
G
A
+
G
W
O
El
d
e
r
l
y
E
C
G
M
u
l
t
i
-
st
a
g
e
8
5
,
7
1
%
S
mal
l
d
a
t
a
se
t
2.
M
E
T
H
O
D
T
h
is
s
tu
d
y
is
a
co
n
tin
u
atio
n
o
f
p
r
ev
io
u
s
r
esear
ch
i
n
v
o
lv
in
g
eld
er
ly
p
ar
ticip
a
n
ts
[
2
3
]
.
Usi
n
g
a
Sp
ar
k
Fu
n
AD8
2
3
2
E
C
G
s
en
s
o
r
an
d
a
n
Ar
d
u
in
o
U
n
o
R
3
,
p
ar
ticip
an
ts
wer
e
ex
p
o
s
ed
t
o
em
o
tio
n
-
in
d
u
cin
g
v
id
eo
s
tim
u
li.
B
ef
o
r
e
s
witch
i
n
g
to
th
e
n
ex
t
e
m
o
tio
n
al
s
tim
u
lu
s
v
id
eo
,
p
ar
ticip
an
ts
wer
e
s
h
o
wn
a
r
elax
atio
n
v
id
eo
to
h
el
p
r
esto
r
e
th
eir
em
o
tio
n
al
s
tate
to
a
n
eu
tr
al
b
aselin
e.
T
h
e
E
C
G
s
ig
n
als
ac
q
u
ir
ed
f
r
o
m
th
e
Sp
a
r
k
Fu
n
AD8
2
3
2
s
en
s
o
r
wer
e
r
ec
o
r
d
e
d
an
d
s
to
r
ed
o
n
a
co
m
p
u
ter
in
C
SV
f
o
r
m
at.
Fig
u
r
e
1
illu
s
tr
ates
th
e
o
v
er
all
m
eth
o
d
o
l
o
g
ical
f
r
am
ew
o
r
k
o
f
th
e
p
r
o
p
o
s
ed
s
tu
d
y
.
T
h
e
d
ataset
u
s
ed
in
th
is
s
tu
d
y
co
n
s
is
ts
o
f
E
C
G
r
ec
o
r
d
in
g
s
o
b
tain
ed
f
r
o
m
eld
er
ly
p
ar
ticip
an
ts
ex
p
o
s
ed
t
o
th
r
ee
em
o
tio
n
al
s
tates:
s
ad
,
s
u
r
p
r
is
ed
,
a
n
d
a
n
g
r
y
.
A
to
tal
o
f
2
7
E
C
G
s
eg
m
en
ts
wer
e
co
llected
,
co
m
p
r
is
in
g
2
1
s
am
p
les f
o
r
tr
ain
in
g
an
d
6
s
am
p
les f
o
r
test
in
g
.
E
ac
h
E
C
G
s
ig
n
al
was p
r
o
ce
s
s
ed
to
ex
tr
ac
t e
ig
h
t
HR
V
-
b
ased
f
ea
tu
r
es,
in
clu
d
i
n
g
Me
an
R
R
,
SDNN,
R
MSS
D,
Me
an
HR
,
STDH
R
,
L
F,
HF,
an
d
L
F/HF
r
atio
.
I
t
is
im
p
o
r
tan
t
to
n
o
te
th
at
th
e
d
ata
s
et
s
ize
i
s
r
elativ
ely
lim
ited
d
u
e
to
th
e
d
if
f
ic
u
lty
o
f
co
llecti
n
g
co
n
tr
o
lled
E
C
G
-
b
ased
em
o
tio
n
al
r
esp
o
n
s
es
f
r
o
m
eld
er
ly
p
ar
ticip
an
ts
.
T
h
er
ef
o
r
e,
th
is
s
tu
d
y
f
o
c
u
s
es
n
o
t
o
n
ly
o
n
class
if
icatio
n
p
er
f
o
r
m
an
ce
b
u
t
also
o
n
o
p
tim
izatio
n
b
eh
a
v
io
r
,
co
n
v
er
g
en
ce
s
tab
ilit
y
,
an
d
r
e
p
r
o
d
u
cib
ili
ty
ac
r
o
s
s
m
u
ltip
le
in
d
ep
en
d
en
t
r
u
n
s
,
wh
ic
h
ar
e
m
o
r
e
s
u
itab
le
e
v
alu
atio
n
cr
it
er
ia
f
o
r
s
m
all
-
s
am
p
le
b
io
m
ed
ical
d
atasets
.
T
h
is
ex
p
er
im
en
tal
d
esig
n
is
s
u
p
p
o
r
ted
b
y
p
s
y
c
h
o
p
h
y
s
io
lo
g
ical
f
in
d
in
g
s
in
d
icatin
g
t
h
at
p
o
s
itiv
e
an
d
n
e
g
ativ
e
af
f
ec
tiv
e
s
tates
s
ig
n
if
ican
tly
i
n
f
lu
en
ce
ca
r
d
i
o
v
ascu
lar
r
esp
o
n
s
es
an
d
a
u
to
n
o
m
ic
n
er
v
o
u
s
s
y
s
tem
r
eg
u
latio
n
[
2
4
]
,
[
2
5
]
.
Fig
u
r
e
2
p
r
esen
ts
th
e
d
ata
ac
q
u
is
itio
n
p
r
o
ce
s
s
f
r
o
m
eld
er
ly
p
a
r
ticip
an
ts
.
Fig
u
r
e
1
.
Ov
e
r
all
m
eth
o
d
o
lo
g
i
ca
l f
r
am
ewo
r
k
o
f
t
h
e
p
r
o
p
o
s
ed
s
tu
d
y
Fig
u
r
e
2
.
Data
co
llectio
n
f
r
o
m
eld
er
ly
p
a
r
ticip
an
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
P
erfo
r
ma
n
ce
a
n
a
lysi
s
o
f sin
g
le
a
n
d
mu
lti
-
s
ta
g
e
meta
h
eu
r
is
tic
o
p
timiz
a
tio
n
o
n
…
(
Gio
va
n
n
i D
ima
s
P
r
en
a
ta
)
1565
2
.
1
.
E
CG
f
ea
t
ure
ex
t
ra
c
t
io
n
B
ef
o
r
e
co
n
d
u
ctin
g
tr
ain
in
g
an
d
test
in
g
u
s
in
g
th
e
f
iv
e
DFFNN
-
b
ased
m
o
d
els
(
Pu
r
e
DFFNN,
DFFNN+G
A,
DFFNN+G
W
O
,
DFFNN+PSO,
an
d
DFFNN
+G
A+
GW
O)
,
a
f
ea
tu
r
e
e
x
tr
ac
tio
n
s
tag
e
was
f
i
r
s
t
p
er
f
o
r
m
ed
o
n
t
h
e
E
C
G
s
ig
n
als
b
ased
o
n
h
ea
r
t
r
ate
v
ar
iab
ilit
y
(
HR
V)
.
T
h
e
f
ea
tu
r
e
e
x
tr
ac
tio
n
p
r
o
ce
s
s
b
eg
an
b
y
r
ea
d
in
g
E
C
G
d
ata
f
r
o
m
C
SV
f
iles
f
o
r
m
atted
as
tim
e;
_
.
T
h
e
ex
tr
ac
ted
v
alu
es
wer
e
s
to
r
e
d
in
th
e
ar
r
ay
s
[
]
an
d
[
]
,
with
a
m
ax
im
u
m
ca
p
ac
ity
o
f
1
0
,
0
0
0
s
am
p
les.
Af
ter
lo
ad
in
g
th
e
d
ata,
th
e
s
y
s
tem
co
m
p
u
ted
t
h
e
m
ea
n
an
d
s
tan
d
ar
d
d
e
v
iatio
n
o
f
th
e
E
C
G
s
ig
n
al
to
estab
lis
h
an
au
to
m
atic
R
-
p
ea
k
d
etec
tio
n
th
r
esh
o
ld
,
d
e
f
in
ed
as
ℎ
ℎ
=
+
1
.
5
×
.
T
h
is
th
r
esh
o
ld
was
ap
p
lied
to
e
n
s
u
r
e
th
at
th
e
d
e
tecte
d
p
ea
k
s
co
r
r
esp
o
n
d
e
d
to
d
o
m
in
an
t
R
-
wav
es
r
ath
e
r
th
an
n
o
is
e
o
r
m
in
o
r
f
lu
ctu
atio
n
s
.
Fig
u
r
e
3
s
h
o
ws t
h
e
E
C
G
f
ea
tu
r
e
ex
tr
ac
tio
n
wo
r
k
f
lo
w.
Fig
u
r
e
3
.
E
C
G
f
ea
tu
r
e
ex
tr
ac
ti
o
n
f
lo
wch
a
r
t
R
-
wav
e
(
R
-
p
ea
k
)
d
etec
tio
n
w
as
th
en
p
er
f
o
r
m
e
d
u
s
in
g
a
lo
c
al
m
ax
im
u
m
cr
iter
io
n
,
wh
er
e
a
s
am
p
le
at
in
d
ex
was
co
n
s
id
er
ed
an
R
-
p
ea
k
if
its
am
p
litu
d
e
ex
c
ee
d
ed
b
o
t
h
n
eig
h
b
o
r
i
n
g
s
am
p
les
(
[
]
>
[
−
1
]
an
d
[
]
>
[
+
1
]
)
an
d
s
u
r
p
ass
ed
th
e
p
r
ed
eter
m
in
e
d
th
r
esh
o
ld
.
T
o
p
r
ev
e
n
t
m
u
ltip
le
d
etec
tio
n
s
ca
u
s
ed
b
y
clo
s
ely
s
p
ac
e
d
p
ea
k
s
,
a
m
in
im
u
m
s
ep
ar
atio
n
o
f
1
0
0
s
am
p
les
b
etwe
en
co
n
s
ec
u
tiv
e
p
ea
k
s
was
en
f
o
r
c
ed
.
T
h
e
d
etec
ted
R
-
p
ea
k
tim
estam
p
s
wer
e
s
to
r
ed
in
t
h
e
ar
r
a
y
[
]
.
B
ased
o
n
th
ese
tim
estam
p
s
,
R
R
in
ter
v
als we
r
e
co
m
p
u
ted
as th
e
tim
e
d
if
f
er
en
ce
s
b
etwe
en
s
u
cc
ess
iv
e
R
-
p
ea
k
s
.
I
n
th
is
im
p
lem
en
tatio
n
,
th
e
E
C
G
s
am
p
lin
g
f
r
e
q
u
en
c
y
was
s
et
to
f
s
=
2
5
0
Hz,
an
d
R
R
in
ter
v
als
w
er
e
ca
lcu
lated
u
s
in
g
(
)
=
(
(
+
1
)
−
(
)
)
/
in
s
ec
o
n
d
s
.
Hea
r
t
r
ate
(
HR
)
was
s
u
b
s
eq
u
en
tly
d
er
iv
e
d
f
r
o
m
th
e
R
R
in
ter
v
als u
s
in
g
(
)
=
60
/
(
)
in
b
ea
ts
p
e
r
m
in
u
te
(
b
p
m
)
.
On
c
e
t
h
e
R
R
i
n
t
er
v
a
ls
a
n
d
H
R
v
al
u
es
we
r
e
o
b
ta
in
e
d
,
HR
V
ti
m
e
-
d
o
m
ai
n
f
e
at
u
r
es
we
r
e
e
x
t
r
a
cte
d
.
T
h
ese
i
n
cl
u
d
e
d
Me
a
n
R
R
,
r
e
p
r
ese
n
ti
n
g
t
h
e
av
er
a
g
e
R
R
i
n
t
er
v
al
;
SDN
N,
t
h
e
s
ta
n
d
a
r
d
d
e
v
ia
tio
n
o
f
R
R
in
te
r
v
als
r
e
f
le
cti
n
g
o
v
e
r
a
ll
h
ea
r
t
r
ate
v
ar
i
a
b
ili
ty
;
an
d
R
MSSD
,
c
o
m
p
u
te
d
as
t
h
e
r
o
o
t
m
e
a
n
s
q
u
a
r
e
o
f
s
u
cc
ess
iv
e
R
R
d
i
f
f
er
e
n
ce
s
t
o
ca
p
t
u
r
e
s
h
o
r
t
-
te
r
m
b
e
at
-
to
-
b
e
at
v
ar
i
a
b
ili
ty
.
I
n
a
d
d
iti
o
n
,
Me
a
n
HR
a
n
d
S
T
DHR
we
r
e
ca
lc
u
la
te
d
as
th
e
m
ea
n
an
d
s
ta
n
d
a
r
d
d
ev
iat
io
n
o
f
t
h
e
h
ea
r
t
r
at
e
s
eq
u
e
n
ce
,
r
es
p
e
cti
v
el
y
.
T
o
g
e
th
e
r
,
th
ese
f
i
v
e
ti
m
e
-
d
o
m
a
in
f
e
at
u
r
es
co
n
s
t
it
u
te
k
e
y
HR
V
d
esc
r
i
p
t
o
r
s
a
n
d
p
r
o
v
i
d
e
d
is
cr
im
i
n
at
iv
e
in
f
o
r
m
ati
o
n
ac
r
o
s
s
e
m
o
t
io
n
al
c
o
n
d
iti
o
n
s
.
I
n
ad
d
itio
n
to
tim
e
-
d
o
m
ain
an
aly
s
is
,
f
r
eq
u
en
c
y
-
d
o
m
ain
HR
V
f
ea
tu
r
es
wer
e
e
x
tr
ac
ted
u
s
in
g
th
e
f
ast
Fo
u
r
ier
t
r
an
s
f
o
r
m
(
FF
T
)
.
T
h
e
R
R
in
ter
v
al
s
eq
u
en
ce
was
f
ir
s
t
d
etr
en
d
ed
b
y
r
em
o
v
in
g
its
m
ea
n
co
m
p
o
n
en
t
th
r
o
u
g
h
th
e
o
p
er
atio
n
R
R
-
M
ea
n
R
R
,
th
en
r
ep
r
esen
ted
as
a
s
ig
n
al
v
ec
to
r
o
f
len
g
t
h
N
=
2
5
6
(
a
p
o
wer
o
f
two
)
f
o
r
FF
T
co
m
p
u
tatio
n
,
with
z
er
o
-
p
a
d
d
in
g
ap
p
lied
wh
en
th
e
n
u
m
b
er
o
f
R
R
in
ter
v
als
w
as
less
th
an
N.
T
h
e
r
esu
ltin
g
FF
T
s
p
ec
tr
u
m
was
u
s
ed
to
esti
m
ate
s
p
ec
tr
al
p
o
wer
with
in
two
p
r
in
cip
al
HR
V
f
r
e
q
u
en
cy
b
an
d
s
:
lo
w
f
r
eq
u
e
n
cy
(
L
F)
in
th
e
r
a
n
g
e
o
f
0
.
0
4
-
0
.
1
5
Hz
an
d
h
i
g
h
f
r
eq
u
e
n
cy
(
HF)
in
th
e
r
an
g
e
o
f
0
.
1
5
-
0
.
4
0
Hz.
T
h
e
L
F/HF
r
atio
was
th
en
ca
lcu
lated
to
r
ef
lect
th
e
b
alan
ce
b
etwe
en
s
y
m
p
ath
etic
a
n
d
p
ar
asy
m
p
ath
etic
n
er
v
o
u
s
s
y
s
tem
ac
tiv
ity
.
I
n
th
e
p
r
o
g
r
am
im
p
lem
e
n
tatio
n
,
th
e
f
r
eq
u
e
n
cy
r
eso
lu
tio
n
was
d
eter
m
in
ed
u
s
in
g
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
1
6
,
No
.
3
,
J
u
n
e
20
2
6
:
1
5
6
2
-
1
5
7
5
1566
_
=
4
Hz
an
d
=
_
/
.
C
o
n
s
eq
u
en
tly
,
th
e
f
ea
tu
r
e
ex
tr
ac
tio
n
s
tag
e
p
r
o
d
u
ce
d
a
to
tal
o
f
eig
h
t
HR
V
f
ea
tu
r
es,
co
m
p
r
is
in
g
f
iv
e
tim
e
-
d
o
m
ai
n
f
ea
tu
r
es
(
Me
an
R
R
,
SDNN,
R
MSS
D,
Me
an
HR
,
STDH
R
)
an
d
th
r
ee
f
r
eq
u
e
n
cy
-
d
o
m
ain
f
ea
tu
r
es
(
L
F,
HF,
L
F/HF
)
,
wh
ich
wer
e
s
u
b
s
eq
u
e
n
tly
u
s
ed
as
tr
ain
in
g
an
d
test
in
g
in
p
u
ts
f
o
r
all
DFFNN m
o
d
els ev
alu
ated
in
t
h
is
s
tu
d
y
.
2
.
2
.
P
ure
DF
F
NN
I
n
th
is
s
tu
d
y
,
a
p
u
r
e
d
ee
p
f
ee
d
f
o
r
war
d
n
eu
r
al
n
etwo
r
k
(
Pu
r
e
DFFN
N)
is
em
p
lo
y
ed
as
th
e
p
r
im
ar
y
b
aselin
e
m
o
d
el
f
o
r
E
C
G
-
b
ase
d
em
o
tio
n
class
if
icatio
n
with
o
u
t
in
co
r
p
o
r
atin
g
a
n
y
m
etah
e
u
r
is
tic
o
p
tim
izatio
n
.
T
h
e
in
p
u
t
d
ata
co
n
s
is
t
o
f
ei
g
h
t
E
C
G/HR
V
-
d
er
iv
ed
f
ea
tu
r
e
s
,
o
r
g
an
ize
d
in
to
an
8
×2
7
m
a
tr
ix
(
eig
h
t
f
ea
tu
r
es
ac
r
o
s
s
2
7
s
am
p
les).
Prio
r
to
b
ein
g
p
r
o
ce
s
s
ed
b
y
th
e
n
e
u
r
a
l
n
etwo
r
k
,
r
o
w
-
wis
e
Min
-
Ma
x
n
o
r
m
aliza
tio
n
is
ap
p
lied
to
ea
c
h
f
ea
tu
r
e,
s
ca
lin
g
all
v
alu
es
in
to
th
e
r
a
n
g
e
o
f
0
to
1
.
T
h
is
n
o
r
m
aliza
tio
n
s
te
p
im
p
r
o
v
es
tr
ain
in
g
s
tab
ilit
y
an
d
p
r
ev
e
n
ts
f
ea
tu
r
es
with
lar
g
er
m
ag
n
itu
d
es f
r
o
m
d
o
m
in
atin
g
th
e
lear
n
in
g
p
r
o
ce
s
s
.
T
h
e
d
ataset
in
th
e
p
r
o
g
r
am
is
d
iv
id
ed
in
to
2
1
tr
ain
in
g
s
am
p
l
es
an
d
s
ix
test
in
g
s
am
p
les.
T
h
e
tr
ain
in
g
tar
g
ets
ar
e
n
u
m
e
r
ically
d
e
f
in
ed
in
th
e
ar
r
a
y
[
9
]
,
w
h
er
e
0
.
1
r
e
p
r
esen
ts
th
e
Sad
class
,
0
.
5
r
e
p
r
esen
ts
th
e
Su
r
p
r
is
e
class
,
an
d
0
.
9
r
ep
r
e
s
en
ts
th
e
an
g
r
y
class
.
A
s
ig
m
o
id
ac
tiv
atio
n
f
u
n
ctio
n
is
u
s
ed
f
o
r
all
n
eu
r
o
n
s
,
en
s
u
r
in
g
th
at
th
e
n
etwo
r
k
o
u
tp
u
ts
r
em
ain
with
in
th
e
in
ter
v
al
[
0
,
1
]
.
B
ased
o
n
th
e
im
p
lem
en
ted
co
d
e,
t
h
e
n
etwo
r
k
ar
c
h
itectu
r
e
co
n
s
is
ts
o
f
m
u
ltip
le
h
i
d
d
en
la
y
er
s
with
th
e
co
n
f
i
g
u
r
atio
n
8
in
p
u
ts
→
2
n
eu
r
o
n
s
(
L
ay
e
r
1
)
→
2
n
eu
r
o
n
s
(
L
a
y
er
2
)
→
2
n
eu
r
o
n
s
(
L
a
y
er
3
)
→
1
o
u
t
p
u
t
n
eu
r
o
n
,
r
esu
ltin
g
in
a
t
o
tal
o
f
3
3
tr
ain
a
b
le
weig
h
t
p
ar
am
eter
s
.
T
h
e
in
itial
weig
h
ts
ar
e
r
an
d
o
m
ly
g
e
n
er
ated
f
r
o
m
a
u
n
if
o
r
m
d
is
tr
ib
u
tio
n
with
in
th
e
r
a
n
g
e
o
f
[
-
1
,
1
]
.
Fig
u
r
e
4
illu
s
tr
ates th
e
tr
ain
in
g
m
ec
h
a
n
is
m
s
o
f
Pu
r
e
DFFNN m
o
d
el.
Fig
u
r
e
4
.
Pu
r
e
DFFNN f
lo
wch
ar
t
T
r
ain
in
g
is
p
er
f
o
r
m
e
d
u
s
in
g
a
s
tan
d
ar
d
f
ee
d
f
o
r
wa
r
d
-
b
ac
k
p
r
o
p
ag
atio
n
m
ec
h
an
is
m
d
r
iv
en
b
y
o
u
tp
u
t
er
r
o
r
m
in
im
izatio
n
.
Fo
r
ea
c
h
i
ter
atio
n
,
all
2
1
tr
ai
n
in
g
s
am
p
les
ar
e
f
o
r
war
d
-
p
r
o
p
a
g
ated
t
o
p
r
o
d
u
ce
th
e
n
etwo
r
k
o
u
tp
u
t
_
,
an
d
th
e
e
r
r
o
r
is
co
m
p
u
ted
u
s
in
g
=
[
]
−
_
.
T
h
is
er
r
o
r
is
th
en
u
s
ed
to
ca
lc
u
late
g
r
ad
ien
ts
an
d
u
p
d
ate
th
e
weig
h
ts
s
eq
u
en
tially
f
r
o
m
th
e
o
u
tp
u
t
lay
er
b
ac
k
to
th
e
i
n
p
u
t
lay
er
u
s
in
g
th
e
d
er
iv
ativ
e
o
f
th
e
s
ig
m
o
id
f
u
n
ctio
n
.
T
h
e
lear
n
in
g
p
r
o
ce
s
s
ad
o
p
ts
a
f
ix
e
d
lear
n
in
g
r
ate
(
L
R
)
o
f
0
.
7
,
wh
ile
t
h
e
o
b
jectiv
e
f
u
n
ctio
n
to
b
e
m
in
i
m
ized
is
th
e
MSE
co
m
p
u
ted
as
th
e
av
er
ag
e
s
q
u
ar
ed
er
r
o
r
o
v
er
th
e
2
1
tr
ain
in
g
s
am
p
les.
T
h
e
tr
ain
in
g
ter
m
in
a
tes
wh
en
th
e
MSE
f
alls
b
elo
w
th
e
p
r
e
d
ef
in
ed
th
r
esh
o
ld
o
f
0
.
0
0
0
1
,
o
r
wh
en
t
h
e
iter
atio
n
co
u
n
t
r
ea
ch
es
th
e
m
a
x
im
u
m
lim
it
o
f
5
,
0
0
0
,
0
0
0
iter
atio
n
s
.
Fo
r
co
n
v
er
g
en
ce
an
aly
s
is
,
th
e
MSE
v
alu
e
at
ea
ch
iter
atio
n
is
r
ec
o
r
d
ed
in
m
s
e_
lo
g
.
csv
in
t
h
e
f
o
r
m
at
(
ite
r
atio
n
,
MSE
)
.
Af
ter
tr
ain
in
g
,
th
e
m
o
d
el
is
ev
alu
ated
o
n
th
e
co
m
p
lete
d
ataset
o
f
2
7
s
am
p
les
(
2
1
tr
ain
in
g
s
am
p
les
an
d
s
ix
test
in
g
s
am
p
les)
u
s
in
g
th
e
lear
n
e
d
weig
h
ts
.
T
h
e
f
in
al
s
ig
m
o
id
o
u
tp
u
t
(
Ou
tp
u
t)
is
m
ap
p
ed
i
n
to
em
o
tio
n
lab
els b
ased
o
n
p
r
ed
e
f
in
ed
in
t
er
v
al
th
r
esh
o
ld
s
:
s
ad
if
o
u
t
p
u
t
<
0
.
4
5
,
s
u
r
p
r
is
e
if
0
.
4
5
≤
S0
7
≤
0
.
8
0
,
an
d
a
n
g
r
y
if
o
u
tp
u
t
>
0
.
8
.
C
lass
if
icatio
n
a
cc
u
r
ac
y
is
ca
lcu
lated
b
y
co
m
p
ar
in
g
th
e
p
r
ed
icted
lab
els
ag
ain
s
t
th
e
r
ef
er
en
ce
lab
els
s
to
r
ed
in
2
[
]
,
an
d
th
e
p
er
f
o
r
m
an
ce
is
r
ep
o
r
ted
s
ep
ar
ate
ly
f
o
r
th
e
tr
ain
in
g
s
et
(
2
1
s
am
p
les)
an
d
th
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
P
erfo
r
ma
n
ce
a
n
a
lysi
s
o
f sin
g
le
a
n
d
mu
lti
-
s
ta
g
e
meta
h
eu
r
is
tic
o
p
timiz
a
tio
n
o
n
…
(
Gio
va
n
n
i D
ima
s
P
r
en
a
ta
)
1567
test
in
g
s
et
(
s
ix
s
am
p
les).
C
o
n
s
eq
u
en
tly
,
th
e
Pu
r
e
DFFNN
m
o
d
el
s
er
v
es
as
th
e
k
ey
b
as
elin
e
f
o
r
ass
ess
in
g
p
er
f
o
r
m
an
ce
im
p
r
o
v
em
en
ts
ac
h
iev
ed
th
r
o
u
g
h
m
etah
e
u
r
is
tic
o
p
tim
izatio
n
s
tr
ateg
ies
s
u
ch
a
s
GA,
G
W
O,
PS
O,
an
d
GA+
GW
O
in
th
is
r
esear
ch
.
T
o
en
s
u
r
e
f
air
co
m
p
a
r
is
o
n
ac
r
o
s
s
all
m
o
d
els,
th
e
class
if
icatio
n
d
ec
is
io
n
th
r
esh
o
ld
s
wer
e
s
tan
d
ar
d
ize
d
f
o
r
all
ex
p
e
r
im
en
ts
.
2
.
3
.
DF
F
NN+G
A
I
n
th
e
DFFNN+GA
m
o
d
el,
th
e
GA
is
em
p
lo
y
ed
to
g
lo
b
all
y
o
p
tim
ize
th
e
weig
h
ts
an
d
b
ia
s
es
o
f
th
e
DFFNN,
th
er
eb
y
elim
in
atin
g
th
e
n
ee
d
f
o
r
g
r
ad
ien
t
-
b
ased
b
a
ck
p
r
o
p
ag
atio
n
.
E
ac
h
ca
n
d
id
ate
s
o
lu
tio
n
in
GA
is
r
ep
r
esen
ted
as
a
ch
r
o
m
o
s
o
m
e
co
n
s
is
tin
g
o
f
3
3
g
en
es,
wh
er
e
ea
ch
g
en
e
co
r
r
esp
o
n
d
s
to
a
s
p
ec
if
ic
weig
h
t
o
r
b
ias
p
ar
am
eter
in
th
e
ad
o
p
ted
DFFNN
ar
ch
itectu
r
e.
T
h
e
i
n
itial
p
o
p
u
latio
n
is
g
en
er
ated
with
1
0
ch
r
o
m
o
s
o
m
es
,
an
d
ea
ch
g
en
e
is
r
a
n
d
o
m
l
y
i
n
itialized
with
in
th
e
r
an
g
e
o
f
-
3
0
t
o
3
0
to
en
s
u
r
e
s
u
f
f
icie
n
t
d
iv
er
s
ity
at
th
e
b
eg
in
n
in
g
o
f
th
e
o
p
tim
izatio
n
p
r
o
ce
s
s
.
T
h
e
m
ain
ex
p
e
r
im
e
n
tal
p
ar
am
ete
r
s
in
clu
d
e
a
cr
o
s
s
o
v
er
p
o
in
t
(
C
P)
o
f
1
0
,
a
m
u
tatio
n
r
ate
o
f
0
.
3
,
a
m
ax
im
u
m
iter
atio
n
lim
it
o
f
5
,
0
0
0
,
0
0
0
,
an
d
an
MSE
th
r
esh
o
ld
o
f
0
.
0
0
0
1
as
t
h
e
s
to
p
p
in
g
cr
iter
io
n
.
T
h
e
q
u
ality
o
f
ea
ch
c
h
r
o
m
o
s
o
m
e
is
ev
alu
ated
u
s
in
g
a
f
itn
ess
f
u
n
ctio
n
b
ased
o
n
MSE
.
Sp
ec
if
ically
,
th
e
3
3
g
e
n
es
o
f
a
ch
r
o
m
o
s
o
m
e
ar
e
ass
ig
n
ed
as
th
e
DFFNN
p
ar
am
eter
s
,
an
d
th
e
n
etwo
r
k
p
er
f
o
r
m
s
f
o
r
war
d
p
r
o
p
a
g
atio
n
o
v
er
all
2
1
tr
ain
in
g
s
am
p
les.
Fo
r
ea
c
h
tr
ain
i
n
g
s
am
p
le,
th
e
n
etwo
r
k
o
u
t
p
u
t
is
c
o
m
p
u
ted
u
s
in
g
th
e
s
ig
m
o
id
ac
tiv
atio
n
f
u
n
ctio
n
at
ev
er
y
n
eu
r
o
n
.
T
h
e
er
r
o
r
i
s
th
en
ca
lcu
lated
as
th
e
d
if
f
er
en
ce
b
etwe
en
th
e
p
r
ed
ef
in
e
d
tar
g
et
v
alu
e
a
n
d
th
e
n
etwo
r
k
o
u
tp
u
t.
T
h
e
o
v
er
all
MSE
is
o
b
tain
ed
b
y
av
er
a
g
in
g
th
e
s
q
u
ar
ed
er
r
o
r
s
ac
r
o
s
s
all
tr
ain
in
g
s
am
p
les,
an
d
th
is
v
al
u
e
is
u
s
ed
as
th
e
f
itn
ess
s
co
r
e,
wh
er
e
a
s
m
aller
MS
E
in
d
icate
s
a
b
etter
ch
r
o
m
o
s
o
m
e.
Fig
u
r
e
5
illu
s
tr
ates th
e
tr
ain
in
g
m
ec
h
a
n
is
m
s
o
f
Pu
r
e
DFFNN+GA
m
o
d
el.
Fig
u
r
e
5
.
Pu
r
e
DFFNN+GA
f
l
o
wch
ar
t
Af
ter
ev
alu
atin
g
all
ch
r
o
m
o
s
o
m
es,
th
e
p
ar
en
t selec
tio
n
s
tag
e
is
p
er
f
o
r
m
ed
.
Par
en
t
-
1
is
s
elec
ted
as th
e
ch
r
o
m
o
s
o
m
e
with
th
e
lo
west
MSE
in
th
e
p
o
p
u
latio
n
,
wh
ile
Par
en
t
-
2
is
c
h
o
s
en
f
r
o
m
o
th
er
h
ig
h
-
q
u
ality
ch
r
o
m
o
s
o
m
es
th
at
ex
h
ib
it
an
MSE
d
if
f
er
en
ce
o
f
at
least
≥
0
.
0
2
c
o
m
p
ar
e
d
to
Par
en
t
-
1
.
T
h
is
s
elec
tio
n
s
tr
ateg
y
is
d
esig
n
ed
to
m
ain
tain
p
o
p
u
la
tio
n
d
iv
er
s
ity
an
d
r
ed
u
ce
th
e
r
i
s
k
o
f
p
r
em
atu
r
e
co
n
v
er
g
en
ce
.
I
f
n
o
ch
r
o
m
o
s
o
m
e
s
atis
f
ies th
i
s
d
if
f
er
en
ce
cr
iter
i
o
n
,
Par
en
t
-
2
is
ass
ig
n
ed
as th
e
s
ec
o
n
d
-
b
est ch
r
o
m
o
s
o
m
e
in
t
h
e
p
o
p
u
latio
n
.
T
h
e
n
ex
t
s
tep
a
p
p
lies
s
in
g
le
-
p
o
in
t
cr
o
s
s
o
v
er
b
ased
o
n
th
e
p
r
ed
ef
in
ed
C
P
v
alu
e.
Gen
es
lo
c
ated
b
ef
o
r
e
th
e
cr
o
s
s
o
v
er
p
o
in
t
ar
e
in
h
er
i
ted
d
ir
ec
tly
f
r
o
m
th
eir
r
esp
ec
tiv
e
p
ar
en
ts
,
wh
er
ea
s
g
en
es
af
ter
th
e
cr
o
s
s
o
v
er
p
o
in
t
ar
e
ex
ch
a
n
g
ed
to
p
r
o
d
u
ce
two
o
f
f
s
p
r
in
g
ch
r
o
m
o
s
o
m
e
s
.
Su
b
s
eq
u
en
tly
,
a
m
u
tatio
n
p
r
o
ce
s
s
is
ap
p
lied
to
ea
ch
g
en
e
with
a
p
r
o
b
ab
ilit
y
o
f
0
.
3
.
W
h
en
m
u
tatio
n
o
cc
u
r
s
,
t
h
e
g
en
e
v
alu
e
is
r
ep
lace
d
b
y
a
n
ew
r
an
d
o
m
v
alu
e
with
in
th
e
r
a
n
g
e
-
3
0
to
3
0
;
o
th
er
wis
e,
th
e
g
e
n
e
r
em
ai
n
s
u
n
ch
an
g
e
d
f
r
o
m
th
e
cr
o
s
s
o
v
er
o
u
tco
m
e
.
T
h
e
two
m
u
tated
o
f
f
s
p
r
in
g
ar
e
th
e
n
r
e
-
ev
alu
ated
u
s
in
g
th
e
s
am
e
f
itn
ess
p
r
o
ce
d
u
r
e
in
v
o
lv
in
g
f
o
r
w
ar
d
p
r
o
p
ag
atio
n
an
d
MSE
co
m
p
u
tatio
n
.
T
o
u
p
d
ate
th
e
p
o
p
u
latio
n
,
a
r
eg
en
er
atio
n
m
ec
h
a
n
is
m
is
im
p
lem
en
ted
b
y
r
ep
lacin
g
th
e
t
wo
wo
r
s
t
-
p
er
f
o
r
m
in
g
c
h
r
o
m
o
s
o
m
es
(
h
ig
h
est
MSE
)
with
th
e
two
n
ewly
g
e
n
er
ated
m
u
tan
t
c
h
r
o
m
o
s
o
m
es.
T
h
is
ev
o
lu
tio
n
ar
y
c
y
cle
co
n
tin
u
es
iter
ativ
ely
u
n
til
t
h
e
b
est
p
o
p
u
latio
n
MSE
f
alls
b
elo
w
th
e
d
ef
in
ed
th
r
esh
o
ld
o
r
th
e
iter
atio
n
co
u
n
t
r
ea
ch
es
th
e
m
ax
im
u
m
lim
it.
Du
r
i
n
g
th
e
o
p
tim
izatio
n
p
r
o
ce
s
s
,
th
e
b
est
MSE
v
alu
e
at
ea
ch
iter
atio
n
is
r
ec
o
r
d
e
d
in
to
mse_
lo
g
.
csv
,
en
ab
lin
g
co
n
v
er
g
e
n
ce
an
aly
s
is
an
d
p
er
f
o
r
m
a
n
ce
m
o
n
ito
r
in
g
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
1
6
,
No
.
3
,
J
u
n
e
20
2
6
:
1
5
6
2
-
1
5
7
5
1568
Af
ter
th
e
o
p
tim
izatio
n
co
n
cl
u
d
es,
th
e
ch
r
o
m
o
s
o
m
e
with
t
h
e
lo
west
MSE
is
s
elec
ted
as
th
e
f
in
al
o
p
tim
al
s
o
lu
tio
n
,
a
n
d
its
g
en
es
ar
e
u
s
ed
as
th
e
f
in
al
DFFNN
weig
h
ts
f
o
r
class
if
icatio
n
.
T
h
e
m
o
d
el
is
th
e
n
test
ed
o
n
b
o
th
th
e
tr
ain
in
g
a
n
d
test
in
g
d
atasets
,
wh
er
e
th
e
f
in
al
n
etwo
r
k
o
u
tp
u
t
is
m
a
p
p
ed
in
to
em
o
tio
n
class
es
u
s
in
g
p
r
ed
ef
in
e
d
d
ec
is
io
n
b
o
u
n
d
ar
ies:
Sad
if
o
u
t
p
u
t <
0
.
4
5
,
Su
r
p
r
is
e
if
0
.
4
5
≤
o
u
tp
u
t
≤
0
.
8
,
an
d
An
g
r
y
if
o
u
tp
u
t
>
0
.
8
.
C
lass
if
icatio
n
ac
cu
r
ac
y
is
co
m
p
u
ted
b
y
c
o
m
p
ar
in
g
th
e
p
r
e
d
icted
em
o
tio
n
lab
els
with
th
e
g
r
o
u
n
d
-
tr
u
th
lab
els in
b
o
th
tr
ai
n
in
g
an
d
test
in
g
s
ets.
2
.
4
.
DF
F
NN+G
WO
I
n
th
e
DFFNN+GWO
ap
p
r
o
a
ch
,
th
e
DFFNN
tr
ain
in
g
p
r
o
c
ess
d
o
es
n
o
t
r
ely
o
n
b
ac
k
p
r
o
p
ag
atio
n
.
I
n
s
tead
,
all
n
etwo
r
k
weig
h
ts
an
d
b
iases
ar
e
o
p
tim
ized
u
s
in
g
th
e
GW
O
.
I
n
th
is
im
p
lem
en
tatio
n
,
ea
ch
wo
lf
r
ep
r
esen
ts
a
ca
n
d
id
ate
s
o
lu
ti
o
n
en
co
d
ed
as
a
3
3
-
d
im
en
s
io
n
al
weig
h
t
v
ec
to
r
(
_
[
33
]
)
,
wh
ich
co
n
tain
s
th
e
co
m
p
lete
s
et
o
f
weig
h
ts
an
d
b
iases
o
f
th
e
a
d
o
p
ted
DFFNN
ar
ch
itectu
r
e
.
T
h
e
E
C
G
in
p
u
t
d
ata
co
n
s
is
t
o
f
8
e
x
tr
ac
ted
f
ea
tu
r
es
,
wh
ich
ar
e
f
ir
s
t
n
o
r
m
alize
d
u
s
in
g
f
ea
tu
r
e
-
wis
e
m
in
-
m
ax
n
o
r
m
aliza
tio
n
s
o
th
at
ea
ch
f
ea
tu
r
e
is
m
a
p
p
ed
in
to
th
e
r
an
g
e
o
f
0
-
1
.
Su
b
s
eq
u
e
n
tly
,
ea
ch
wo
lf
u
n
d
er
g
o
es
f
o
r
wa
r
d
p
r
o
p
a
g
atio
n
o
n
th
e
2
1
tr
ain
in
g
s
am
p
les
to
p
r
o
d
u
c
e
th
e
n
etwo
r
k
o
u
tp
u
t
(
S0
7
)
.
T
h
e
p
r
ed
ictio
n
e
r
r
o
r
is
co
m
p
u
te
d
as
th
e
d
if
f
er
en
ce
b
etwe
en
th
e
tar
g
et
v
alu
e
a
n
d
t
h
e
n
etwo
r
k
o
u
tp
u
t,
d
ef
in
ed
as
=
−
07
.
T
h
e
f
itn
ess
v
alu
e
o
f
ea
ch
w
o
lf
is
th
en
ev
alu
ated
u
s
in
g
th
e
M
SE,
ca
lcu
lated
as
th
e
av
er
ag
e
s
q
u
ar
ed
er
r
o
r
o
v
er
all
tr
ain
in
g
s
am
p
les,
wh
er
e
a
lo
wer
MSE
in
d
icate
s
a
b
etter
s
o
lu
tio
n
.
Fig
u
r
e
6
illu
s
tr
ates
th
e
tr
ain
in
g
m
ec
h
an
is
m
s
o
f
P
u
r
e
DFFNN+GW
O
m
o
d
el.
Fig
u
r
e
6
.
Pu
r
e
DFFNN+GW
O
f
lo
wch
ar
t
At
ea
ch
iter
atio
n
,
th
e
th
r
ee
wo
lv
es
with
th
e
s
m
alles
t
MSE
v
alu
es
ar
e
s
elec
ted
as
th
e
alp
h
a
(
b
est),
b
eta
(
s
ec
o
n
d
b
est),
a
n
d
d
elta
(
th
ir
d
b
est)
wo
lv
es.
T
h
ese
th
r
ee
wo
lv
es
s
er
v
e
as
lead
e
r
s
th
at
g
u
id
e
t
h
e
s
ea
r
ch
p
r
o
ce
s
s
an
d
d
eter
m
in
e
th
e
m
o
v
em
en
t
o
f
th
e
r
em
ai
n
in
g
w
o
lv
es
to
war
d
th
e
o
p
tim
al
s
o
lu
tio
n
.
T
h
e
p
o
s
itio
n
s
(
weig
h
t
v
ec
to
r
s
)
o
f
all
wo
l
v
es
ex
clu
d
in
g
alp
h
a
,
b
eta,
a
n
d
d
e
lta
ar
e
u
p
d
ate
d
u
s
in
g
t
h
e
s
tan
d
ar
d
GW
O
p
o
s
itio
n
u
p
d
ate
m
ec
h
an
is
m
g
o
v
er
n
e
d
b
y
th
e
co
n
tr
o
l p
ar
am
eter
a,
wh
i
ch
d
ec
r
ea
s
es
lin
ea
r
ly
f
r
o
m
2
t
o
0
as
th
e
iter
atio
n
s
p
r
o
g
r
ess
.
Fo
r
ea
ch
weig
h
t
co
m
p
o
n
e
n
t
j,
th
e
d
is
tan
ce
s
to
th
e
alp
h
a,
b
eta,
a
n
d
d
elta
wo
lv
e
s
ar
e
co
m
p
u
ted
,
a
n
d
th
r
ee
ca
n
d
i
d
ate
p
o
s
itio
n
s
(
X1
,
X2
,
an
d
X3
)
ar
e
g
en
er
ated
u
s
in
g
r
an
d
o
m
co
ef
f
icie
n
ts
(
r
1
a
n
d
r
2
)
th
at
p
r
o
d
u
ce
th
e
in
ter
m
ed
iate
p
ar
am
eter
s
A
an
d
C
.
T
h
e
u
p
d
ate
d
weig
h
t
v
alu
e
is
th
en
o
b
tain
ed
as
th
e
av
er
ag
e
o
f
th
ese
th
r
ee
ca
n
d
id
ate
p
o
s
itio
n
s
,
en
a
b
lin
g
ea
ch
wo
lf
to
ad
ju
s
t
its
p
o
s
itio
n
b
ased
o
n
t
h
e
co
m
b
i
n
ed
in
f
l
u
en
ce
o
f
t
h
e
alp
h
a
,
b
eta,
an
d
d
elta
lead
er
s
.
T
h
is
it
er
ativ
e
o
p
tim
izatio
n
c
o
n
tin
u
es
u
n
til
th
e
s
to
p
p
in
g
cr
iter
io
n
is
m
et,
n
am
el
y
wh
e
n
th
e
b
est
MSE
≤
0
.
0
0
1
o
r
wh
e
n
th
e
iter
atio
n
r
ea
ch
es
th
e
m
a
x
im
u
m
lim
it
o
f
5
,
0
0
0
,
0
0
0
iter
atio
n
s
.
Du
r
in
g
t
h
e
o
p
tim
izatio
n
p
r
o
ce
s
s
,
th
e
b
est M
SE
at
ea
ch
iter
atio
n
(
i.e
.
,
th
e
alp
h
a
wo
l
f
’
s
MSE
)
is
r
ec
o
r
d
e
d
in
to
mse_
lo
g
.
csv
to
g
en
er
ate
th
e
co
n
v
er
g
e
n
ce
c
u
r
v
e
f
o
r
p
e
r
f
o
r
m
an
ce
a
n
aly
s
is
.
2
.
5
.
DF
F
NN+PSO
I
n
th
e
DFFNN+PS
O
m
eth
o
d
,
th
e
tr
ain
in
g
p
r
o
ce
s
s
o
f
th
e
DFFNN
is
p
er
f
o
r
m
ed
b
y
o
p
ti
m
izin
g
th
e
n
etwo
r
k
weig
h
ts
u
s
in
g
PSO
r
ath
er
th
a
n
c
o
n
v
e
n
tio
n
al
b
ac
k
p
r
o
p
ag
atio
n
.
I
n
th
is
f
r
am
ew
o
r
k
,
ea
ch
p
ar
ticle
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
P
erfo
r
ma
n
ce
a
n
a
lysi
s
o
f sin
g
le
a
n
d
mu
lti
-
s
ta
g
e
meta
h
eu
r
is
tic
o
p
timiz
a
tio
n
o
n
…
(
Gio
va
n
n
i D
ima
s
P
r
en
a
ta
)
1569
r
ep
r
esen
ts
a
ca
n
d
id
ate
s
o
lu
tio
n
en
co
d
e
d
as
a
3
3
-
d
im
en
s
io
n
al
DFFNN
we
ig
h
t
v
ec
to
r
(
[
33
]
)
,
in
clu
d
in
g
b
ias
ter
m
s
,
an
d
is
a
s
s
o
ciate
d
with
a
co
r
r
esp
o
n
d
in
g
v
elo
city
v
ec
to
r
(
[
33
]
)
th
at
co
n
tr
o
ls
th
e
u
p
d
ate
s
tep
in
th
e
s
ea
r
ch
s
p
ac
e.
T
h
e
E
C
G
in
p
u
t
d
ata,
co
n
s
is
tin
g
o
f
8
e
x
tr
ac
ted
f
ea
tu
r
es,
a
r
e
f
ir
s
t
n
o
r
m
alize
d
u
s
in
g
f
ea
tu
r
e
-
wis
e
m
in
-
m
ax
n
o
r
m
ali
za
tio
n
(
r
o
w
-
wis
e
in
th
e
co
d
e)
to
s
ca
le
all
v
alu
es
in
to
th
e
r
an
g
e
o
f
0
-
1
,
th
er
eb
y
im
p
r
o
v
in
g
n
u
m
e
r
ical
s
tab
ilit
y
d
u
r
in
g
th
e
o
p
tim
izatio
n
p
r
o
ce
s
s
.
T
h
e
PS
O
s
war
m
is
in
itialized
with
1
0
p
ar
ticles,
wh
er
e
t
h
e
in
itial
wei
g
h
t
p
o
s
itio
n
s
ar
e
r
an
d
o
m
ly
g
e
n
er
ated
with
in
-
1
to
1
,
wh
ile
t
h
e
in
itial
v
elo
cities
ar
e
r
an
d
o
m
ly
ass
ig
n
ed
with
in
-
0
.
1
to
0
.
1
.
T
h
e
o
p
tim
izatio
n
p
ar
am
eter
s
u
s
ed
in
th
e
im
p
lem
e
n
tatio
n
ar
e
c1
=
1
.
5
an
d
c2
=
1
.
5
,
with
a
tar
g
et
er
r
o
r
o
f
MSE
=
0
.
0
0
0
1
an
d
a
m
ax
i
m
u
m
iter
atio
n
lim
it
o
f
5
,
0
0
0
,
0
0
0
.
Fig
u
r
e
7
illu
s
tr
ates th
e
tr
ain
in
g
m
ec
h
an
is
m
s
o
f
Pu
r
e
DFFNN+PS
O
m
o
d
el.
Fig
u
r
e
7
.
Pu
r
e
DFFNN+PS
O
f
lo
wch
ar
t
At
ea
ch
iter
atio
n
,
ev
er
y
p
ar
ticle
is
ev
alu
ated
b
y
ap
p
ly
i
n
g
its
cu
r
r
en
t
weig
h
t
v
ec
t
o
r
th
r
o
u
g
h
a
DFFNN
f
o
r
war
d
p
r
o
p
ag
atio
n
p
r
o
ce
d
u
r
e
o
n
th
e
2
1
tr
ai
n
in
g
s
am
p
l
es
to
p
r
o
d
u
ce
th
e
n
etwo
r
k
o
u
tp
u
t
(
S0
7
)
.
T
h
e
p
r
ed
ictio
n
er
r
o
r
is
co
m
p
u
ted
a
s
E
R
R
=
T
-
S0
7
,
an
d
th
e
p
ar
tic
le’
s
f
itn
ess
is
q
u
an
tifie
d
u
s
in
g
th
e
MSE
av
er
ag
e
d
ac
r
o
s
s
all
tr
ain
in
g
s
am
p
les.
T
h
u
s
,
p
ar
ticles
with
lo
wer
MSE
v
alu
es
in
d
icate
b
ette
r
ca
n
d
i
d
ate
weig
h
t
co
n
f
ig
u
r
atio
n
s
.
On
ce
th
e
MSE
is
o
b
tain
ed
,
th
e
alg
o
r
ith
m
u
p
d
ates
ea
ch
p
ar
ticle’
s
p
er
s
o
n
al
b
est
(
p
_
b
est)
as
th
e
b
est
s
o
lu
tio
n
it
h
as
ac
h
iev
ed
s
o
f
ar
,
an
d
d
eter
m
in
es
th
e
g
lo
b
al
b
est
(
g
_
b
est)
as
th
e
b
est
p
_
b
est
am
o
n
g
all
p
ar
ticles
in
th
e
s
war
m
.
T
h
e
m
ain
PS
O
o
p
tim
izatio
n
s
tep
is
th
en
co
n
d
u
cted
b
y
u
p
d
atin
g
th
e
p
ar
ticle
v
elo
city
an
d
p
o
s
itio
n
b
ased
o
n
th
e
c
o
m
b
in
ed
in
f
lu
en
ce
o
f
th
e
co
g
n
itiv
e
an
d
s
o
cial
co
m
p
o
n
en
ts
.
Sp
ec
if
ically
,
th
e
v
elo
city
is
u
p
d
ated
u
s
in
g
r
a
n
d
o
m
co
ef
f
icien
ts
r
1
a
n
d
r
2
,
an
d
th
e
d
if
f
er
e
n
ce
s
b
etwe
en
th
e
c
u
r
r
en
t
p
o
s
itio
n
an
d
b
o
th
p
_
b
est
an
d
g
_
b
est,
wh
il
e
th
e
n
ew
p
o
s
itio
n
is
o
b
tain
e
d
b
y
ad
d
in
g
t
h
e
u
p
d
ated
v
el
o
city
to
t
h
e
cu
r
r
en
t
p
o
s
itio
n
.
T
h
e
b
est
MSE
v
alu
e
(
g
_
b
est)
at
ea
c
h
iter
atio
n
is
r
ec
o
r
d
e
d
in
t
o
m
s
e_
lo
g
.
c
s
v
to
g
e
n
er
ate
th
e
co
n
v
er
g
en
ce
p
r
o
f
ile.
T
h
e
iter
ativ
e
p
r
o
ce
s
s
is
ter
m
in
ated
o
n
ce
th
e
b
est
MSE
s
atis
f
ie
s
th
e
p
r
ed
ef
in
ed
tar
g
et
th
r
esh
o
ld
o
r
th
e
iter
atio
n
co
u
n
t r
ea
ch
es th
e
m
ax
im
u
m
lim
it.
Af
ter
th
e
o
p
tim
izatio
n
s
tag
e
is
co
m
p
leted
,
th
e
f
in
al
weig
h
t
co
n
f
ig
u
r
atio
n
u
s
ed
f
o
r
class
if
icatio
n
is
tak
en
f
r
o
m
th
e
p
_
b
est
v
e
cto
r
o
f
th
e
g
lo
b
al
b
est
p
ar
ticle,
an
d
th
e
tr
ain
ed
DFFNN
is
ev
alu
ated
o
n
th
e
en
tir
e
d
ataset
o
f
2
7
s
am
p
les
(
2
1
t
r
ain
in
g
s
am
p
les
an
d
6
test
in
g
s
a
m
p
les).
T
h
e
f
i
n
al
n
etwo
r
k
o
u
tp
u
t
S0
7
is
co
n
v
er
ted
in
to
d
is
cr
ete
em
o
tio
n
class
es
b
ased
o
n
p
r
ed
ef
in
e
d
d
ec
is
io
n
r
u
les
im
p
lem
en
ted
in
t
h
e
p
r
o
g
r
am
,
n
am
ely
Sad
if
S0
7
<
0
.
4
5
,
Su
r
p
r
is
ed
if
0
.
4
5
≤
S0
7
≤
0
.
8
0
,
an
d
A
n
g
r
y
if
S0
7
>
0
.
8
0
.
T
h
e
class
if
icatio
n
ac
cu
r
ac
y
is
co
m
p
u
ted
b
y
co
m
p
a
r
in
g
th
e
p
r
ed
icted
e
m
o
tio
n
lab
els with
th
e
co
r
r
esp
o
n
d
in
g
g
r
o
u
n
d
-
tr
u
th
lab
els (
T
2
)
,
s
ep
ar
ately
f
o
r
th
e
tr
ain
in
g
an
d
test
in
g
s
ets.
T
h
is
p
r
o
ce
d
u
r
e
e
n
ab
les
a
co
n
s
is
ten
t
p
er
f
o
r
m
an
ce
ev
alu
atio
n
o
f
th
e
DFFNN+PS
O
m
o
d
el
an
d
s
u
p
p
o
r
ts
a
f
air
co
m
p
ar
is
o
n
with
th
e
o
th
er
m
etah
e
u
r
is
tic
-
b
ased
v
ar
ian
ts
ex
am
in
e
d
in
th
is
s
tu
d
y
.
2
.
6
.
DF
F
NN+G
A+GWO
I
n
th
e
p
r
o
p
o
s
ed
DFFNN+GA
+G
W
O
(
m
u
lti
-
s
tag
e
m
etah
eu
r
is
tic)
ap
p
r
o
ac
h
,
th
e
tr
ain
in
g
p
r
o
ce
s
s
o
f
th
e
DFFNN
is
co
n
d
u
cted
th
r
o
u
g
h
two
s
eq
u
en
tial
o
p
tim
iz
atio
n
s
tag
es.
T
h
e
f
ir
s
t
s
tag
e
em
p
lo
y
s
a
GA
to
id
en
tify
a
h
ig
h
-
q
u
ality
in
itial
s
et
o
f
n
etwo
r
k
weig
h
ts
,
a
n
d
t
h
e
s
ec
o
n
d
s
tag
e
ap
p
lies
th
e
G
W
O
to
f
u
r
th
er
r
ef
in
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
1
6
,
No
.
3
,
J
u
n
e
20
2
6
:
1
5
6
2
-
1
5
7
5
1570
th
e
GA
-
d
er
iv
e
d
s
o
lu
tio
n
.
Simi
lar
to
th
e
o
t
h
er
m
o
d
els,
th
e
E
C
G
in
p
u
t
d
ata
co
n
s
is
t
o
f
eig
h
t
ex
tr
ac
ted
f
ea
tu
r
es,
wh
ich
ar
e
n
o
r
m
alize
d
u
s
in
g
f
ea
tu
r
e
-
wis
e
m
in
-
m
ax
s
ca
lin
g
t
o
en
s
u
r
e
th
at
all
f
ea
tu
r
es
f
all
with
in
th
e
r
an
g
e
o
f
0
-
1
.
T
h
e
DFFNN
co
n
tain
s
3
3
tr
ain
ab
le
weig
h
t
p
ar
am
et
er
s
,
in
clu
d
in
g
b
ias
ter
m
s
;
th
er
ef
o
r
e
,
ea
ch
GA
ch
r
o
m
o
s
o
m
e
an
d
ea
ch
GW
O
wo
lf
r
ep
r
esen
ts
a
3
3
-
d
im
e
n
s
io
n
al
ca
n
d
id
ate
weig
h
t v
ec
to
r
.
Du
r
in
g
th
e
GA
p
h
ase,
an
in
itial
p
o
p
u
latio
n
o
f
1
0
ch
r
o
m
o
s
o
m
es
is
g
en
er
ated
,
wh
er
e
ea
ch
ch
r
o
m
o
s
o
m
e
co
n
tain
s
3
3
wei
g
h
t
g
e
n
es
r
an
d
o
m
ly
i
n
itialized
with
in
th
e
r
an
g
e
o
f
-
3
0
to
3
0
.
E
ac
h
ch
r
o
m
o
s
o
m
e
is
ev
alu
ated
b
y
p
er
f
o
r
m
in
g
f
o
r
war
d
p
r
o
p
ag
atio
n
th
r
o
u
g
h
th
e
DFFNN a
cr
o
s
s
th
e
2
1
tr
ain
in
g
s
am
p
les,
f
o
llo
wed
b
y
co
m
p
u
tin
g
th
e
p
r
ed
ictio
n
er
r
o
r
as
E
R
R
=
T
-
S0
7
.
T
h
e
f
itn
ess
o
f
ea
ch
ch
r
o
m
o
s
o
m
e
is
th
en
q
u
an
tifie
d
u
s
in
g
th
e
MSE
.
T
h
e
s
elec
tio
n
m
ec
h
an
is
m
id
en
tifie
s
th
e
ch
r
o
m
o
s
o
m
e
with
th
e
s
m
allest
MSE
as
Par
en
t
-
1
,
wh
ile
Par
en
t
-
2
is
ch
o
s
en
u
s
in
g
a
r
an
k
in
g
-
b
ased
s
tr
ateg
y
th
at
c
o
n
s
id
er
s
th
e
MSE
d
if
f
er
en
ce
f
r
o
m
th
e
b
est
in
d
iv
id
u
al,
en
s
u
r
in
g
t
h
at
th
e
s
elec
ted
p
ar
en
ts
ar
e
n
o
t
o
v
er
ly
s
im
ilar
.
A
s
in
g
le
-
p
o
in
t
cr
o
s
s
o
v
er
is
th
en
p
er
f
o
r
m
ed
u
s
in
g
a
cr
o
s
s
o
v
er
p
o
in
t
o
f
C
P
=
1
0
to
p
r
o
d
u
ce
two
o
f
f
s
p
r
in
g
.
Mu
tat
io
n
is
s
u
b
s
eq
u
en
tly
ap
p
lied
with
a
p
r
o
b
ab
ilit
y
o
f
0
.
3
,
wh
er
e
s
elec
ted
g
en
es
ar
e
r
ep
lace
d
b
y
n
ewly
g
en
er
ated
r
an
d
o
m
v
al
u
es.
T
h
e
r
esu
ltin
g
m
u
tan
t
o
f
f
s
p
r
i
n
g
ar
e
re
-
ev
alu
ated
i
n
ter
m
s
o
f
MSE
,
an
d
a
r
eg
en
e
r
atio
n
s
tep
is
ca
r
r
ied
o
u
t
b
y
r
ep
lacin
g
th
e
two
wo
r
s
t
-
p
er
f
o
r
m
in
g
ch
r
o
m
o
s
o
m
es
in
th
e
p
o
p
u
latio
n
with
th
e
two
m
u
tan
ts
.
T
h
i
s
GA
ev
o
lu
tio
n
p
r
o
ce
s
s
co
n
tin
u
es
u
n
til
th
e
b
est
MSE
f
alls
b
elo
w
0
.
0
0
0
1
o
r
th
e
m
ax
im
u
m
n
u
m
b
er
o
f
iter
atio
n
s
is
r
ea
ch
ed
,
an
d
th
e
c
o
n
v
e
r
g
en
ce
tr
ajec
to
r
y
is
r
ec
o
r
d
e
d
in
m
s
e_
lo
g
_
GA.
csv
.
Fig
u
r
e
8
illu
s
tr
ates
th
e
tr
ain
in
g
m
ec
h
an
is
m
s
o
f
Pu
r
e
DFFNN+GA
+
GW
O
m
o
d
el.
Fig
u
r
e
8
.
Pu
r
e
DFFNN+GA
+
GW
O
f
lo
wch
ar
t
On
ce
th
e
GA
s
tag
e
c
o
n
v
e
r
g
es
,
th
e
b
est
ch
r
o
m
o
s
o
m
e
is
e
x
tr
ac
ted
as
th
e
o
p
tim
al
GA
wei
g
h
t
v
e
cto
r
an
d
u
s
ed
as
th
e
in
itializatio
n
f
o
r
th
e
s
u
b
s
eq
u
en
t
GW
O
s
tag
e.
I
n
th
e
GW
O
p
h
ase,
a
p
a
ck
o
f
1
0
wo
l
v
es
is
f
o
r
m
ed
,
wh
er
e
t
h
e
f
ir
s
t
wo
lf
(
in
itial
alp
h
a)
is
d
ir
ec
tly
ass
ig
n
ed
th
e
GA
-
b
est
weig
h
ts
,
w
h
ile
th
e
r
em
ain
in
g
wo
lv
es
ar
e
g
en
er
ate
d
b
y
a
d
d
in
g
a
s
m
all
r
an
d
o
m
p
er
t
u
r
b
a
tio
n
(
n
o
is
e)
to
th
e
GA
s
o
lu
t
io
n
.
T
h
is
s
tr
ateg
y
p
r
eser
v
es
th
e
s
tr
o
n
g
s
tar
tin
g
p
o
in
t
f
o
u
n
d
b
y
GA
wh
ile
m
ain
tain
in
g
s
u
f
f
icien
t
d
iv
e
r
s
ity
f
o
r
ex
p
lo
r
atio
n
.
E
ac
h
wo
lf
is
th
en
ev
al
u
ated
u
s
in
g
th
e
s
am
e
DFFNN
f
o
r
war
d
p
r
o
p
ag
atio
n
s
ch
em
e
o
n
th
e
2
1
tr
ain
in
g
s
am
p
les
to
co
m
p
u
te
its
MSE
.
B
ased
o
n
th
e
MSE
r
an
k
in
g
,
th
e
th
r
ee
b
est wo
lv
es a
r
e
s
elec
ted
as a
lp
h
a,
b
eta,
an
d
d
elta.
T
h
e
wo
lf
p
o
s
itio
n
u
p
d
ate
f
o
llo
ws
th
e
s
tan
d
a
r
d
GW
O
m
ec
h
an
i
s
m
co
n
tr
o
lled
b
y
c
o
ef
f
icien
ts
A
an
d
C
,
a
n
d
t
h
e
p
ar
am
eter
a,
w
h
ich
d
ec
r
ea
s
e
s
lin
ea
r
ly
f
r
o
m
2
to
0
as
i
ter
atio
n
s
p
r
o
g
r
ess
.
T
h
is
d
esig
n
e
n
ab
les
wid
er
ex
p
lo
r
atio
n
i
n
ea
r
ly
iter
atio
n
s
an
d
g
r
a
d
u
ally
s
h
if
ts
to
war
d
e
x
p
lo
itatio
n
ar
o
u
n
d
t
h
e
b
est
s
o
lu
tio
n
s
.
T
h
e
GW
O
p
r
o
ce
s
s
ter
m
in
ates
wh
en
th
e
alp
h
a
wo
lf
ac
h
iev
es
an
MSE
≤
0
.
0
0
1
o
r
th
e
m
ax
im
u
m
n
u
m
b
er
o
f
iter
atio
n
s
is
r
ea
ch
ed
,
a
n
d
th
e
o
p
tim
izatio
n
p
r
o
g
r
ess
is
lo
g
g
e
d
in
mse_
lo
g
_
GWO.
csv
.
I
n
th
e
f
in
al
ev
alu
atio
n
s
tag
e,
th
e
em
o
tio
n
class
if
icatio
n
is
p
er
f
o
r
m
ed
u
s
in
g
th
e
b
est
weig
h
t
v
ec
to
r
o
b
tain
ed
f
r
o
m
GW
O,
s
p
ec
if
ic
ally
th
e
weig
h
ts
o
f
th
e
al
p
h
a
wo
lf
in
th
e
f
i
n
al
iter
atio
n
.
T
h
e
tr
ain
ed
m
o
d
el
is
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
P
erfo
r
ma
n
ce
a
n
a
lysi
s
o
f sin
g
le
a
n
d
mu
lti
-
s
ta
g
e
meta
h
eu
r
is
tic
o
p
timiz
a
tio
n
o
n
…
(
Gio
va
n
n
i D
ima
s
P
r
en
a
ta
)
1571
th
en
test
ed
o
n
all
2
7
s
am
p
les
(
2
1
t
r
ain
in
g
s
am
p
les
an
d
6
te
s
tin
g
s
am
p
les)
to
g
en
e
r
ate
th
e
o
u
tp
u
t
v
alu
e
S0
7
.
T
h
e
co
n
tin
u
o
u
s
o
u
tp
u
t
is
s
u
b
s
eq
u
en
tly
m
a
p
p
ed
i
n
to
d
is
c
r
ete
em
o
tio
n
class
es
u
s
in
g
th
e
d
ec
is
io
n
r
u
les
im
p
lem
en
ted
in
th
e
p
r
o
g
r
a
m
,
n
am
ely
Sad
if
S0
7
<
0
.
4
5
,
Su
r
p
r
is
ed
if
0
.
4
5
≤
S0
7
≤
0
.
8
0
,
an
d
An
g
r
y
if
S0
7
>
0
.
8
0
.
C
lass
if
icatio
n
ac
cu
r
ac
y
i
s
co
m
p
u
ted
b
y
co
m
p
ar
in
g
th
e
p
r
ed
icted
em
o
tio
n
lab
el
ag
ai
n
s
t
th
e
g
r
o
u
n
d
-
tr
u
t
h
class
lab
els
(
T
2
)
,
r
ep
o
r
ted
s
ep
ar
ately
f
o
r
th
e
tr
ain
in
g
a
n
d
tes
tin
g
s
ets.
Ov
er
all,
th
e
DFFNN
+G
A+
GW
O
m
o
d
el
r
ep
r
esen
ts
a
m
u
lti
-
s
tag
e
o
p
tim
izatio
n
s
tr
ateg
y
,
wh
er
e
GA
p
r
o
v
id
es
a
r
o
b
u
s
t
g
lo
b
al
s
ea
r
ch
f
o
r
p
r
o
m
is
in
g
in
itial
s
o
lu
tio
n
s
,
an
d
GW
O
f
u
r
th
er
e
n
h
an
ce
s
s
o
lu
tio
n
q
u
ality
th
r
o
u
g
h
f
in
e
-
g
r
ain
ed
ex
p
lo
itatio
n
,
lead
in
g
to
a
m
o
r
e
s
tab
le
co
n
v
er
g
e
n
ce
b
eh
a
v
io
r
a
n
d
im
p
r
o
v
ed
p
r
ed
ictiv
e
p
er
f
o
r
m
an
ce
.
Alth
o
u
g
h
th
e
n
u
m
b
er
o
f
iter
atio
n
s
u
s
ed
in
th
is
s
tu
d
y
is
r
elativ
ely
lar
g
e
,
th
is
d
esig
n
is
in
te
n
tio
n
ally
a
d
o
p
ted
t
o
e
n
s
u
r
e
th
at
all
o
p
tim
izatio
n
alg
o
r
ith
m
s
r
ea
ch
a
s
tab
le
co
n
v
e
r
g
en
ce
r
eg
io
n
.
T
h
is
ap
p
r
o
ac
h
en
a
b
les
a
f
air
co
m
p
ar
is
o
n
o
f
f
i
n
a
l
p
er
f
o
r
m
an
ce
an
d
co
n
v
er
g
en
ce
b
e
h
av
io
r
ac
r
o
s
s
a
ll m
o
d
els.
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
h
is
s
tu
d
y
ev
alu
ates
th
e
p
er
f
o
r
m
an
ce
o
f
a
DFFNN
u
n
d
e
r
v
ar
io
u
s
s
in
g
le
-
s
tag
e
an
d
m
u
lti
-
s
tag
e
m
etah
eu
r
is
tic
o
p
tim
izatio
n
s
tr
ateg
ies
f
o
r
E
C
G
-
b
ased
em
o
tio
n
class
if
icatio
n
.
Fiv
e
m
o
d
els
wer
e
co
m
p
ar
ed
,
n
am
ely
Pu
r
e
DFFNN,
DFFNN
with
GA
,
DFFNN
with
G
W
O
,
DFFNN
with
PSO
,
an
d
t
h
e
m
u
lti
-
s
tag
e
m
o
d
el
DFFNN+G
A+
GW
O
.
All
m
o
d
els
wer
e
as
s
es
s
ed
u
s
in
g
s
ix
in
d
ep
en
d
e
n
t
ex
p
er
im
e
n
tal
tr
ials
to
en
s
u
r
e
th
at
th
e
r
esu
lts
wer
e
n
o
t
d
ep
e
n
d
en
t
o
n
a
s
in
g
le
i
n
itializatio
n
s
ettin
g
b
u
t
in
s
tead
r
ef
lecte
d
th
e
g
en
er
al
b
eh
av
i
o
r
o
f
ea
c
h
o
p
tim
izatio
n
ap
p
r
o
ac
h
.
Fro
m
T
ab
le
2
b
ased
o
n
t
h
e
r
e
s
u
lts
s
u
m
m
ar
ized
,
Pu
r
e
DFFNN
ex
h
ib
ited
th
e
lo
west
p
er
f
o
r
m
an
ce
i
n
ter
m
s
o
f
b
o
t
h
tr
ain
in
g
an
d
test
in
g
ac
cu
r
ac
y
,
s
h
o
win
g
a
r
elativ
ely
h
ig
h
f
in
al
MSE
an
d
a
lar
g
e
v
ar
iatio
n
ac
r
o
s
s
r
ep
ea
ted
tr
ials
.
T
h
ese
f
in
d
in
g
s
in
d
icate
th
at
g
r
ad
ien
t
-
b
ased
lear
n
in
g
al
o
n
e
is
in
s
u
f
f
icien
t
t
o
ef
f
ec
tiv
ely
h
an
d
le
th
e
co
m
p
lex
ity
o
f
E
C
G
s
ig
n
als,
wh
ich
ar
e
in
h
er
e
n
tly
n
o
n
lin
ea
r
,
n
o
n
-
s
tatio
n
ar
y
,
a
n
d
h
i
g
h
l
y
co
n
tam
in
ated
b
y
n
o
is
e.
I
n
Pu
r
e
DFFNN,
th
e
o
p
tim
izatio
n
p
r
o
ce
s
s
ten
d
s
to
b
ec
o
m
e
tr
ap
p
e
d
in
lo
ca
l
m
in
im
u
m
,
a
n
d
its
p
er
f
o
r
m
an
ce
is
s
tr
o
n
g
l
y
d
ep
e
n
d
en
t o
n
th
e
in
itial we
ig
h
t c
o
n
f
ig
u
r
atio
n
.
T
ab
le
2
.
Per
f
o
r
m
an
ce
co
m
p
a
r
is
o
n
o
f
f
i
v
e
DFFNN m
o
d
els ac
r
o
s
s
s
ix
ex
p
er
im
en
tal
r
u
n
s
(
f
in
al
MSE
,
tr
ain
in
g
ac
cu
r
ac
y
,
an
d
test
in
g
ac
cu
r
ac
y
)
Te
st
P
u
r
e
D
F
F
N
N
D
F
F
N
N
+
G
A
F
i
n
a
l
M
S
E
Tr
a
i
n
i
n
g
A
c
c
u
r
a
c
y
(
%)
Te
st
A
c
c
u
r
a
c
y
(
%)
F
i
n
a
l
M
S
E
Tr
a
i
n
i
n
g
A
c
c
u
r
a
c
y
(
%)
Te
st
A
c
c
u
r
a
c
y
(
%)
1
0
.
0
8
1
9
1
6
1
.
9
0
3
3
.
3
3
0
.
0
4
6
2
1
7
1
.
4
3
3
3
.
3
3
2
0
.
0
7
4
6
2
6
6
.
6
7
5
0
.
0
0
0
.
0
4
3
8
8
6
6
.
6
7
5
0
.
0
0
3
0
.
0
8
9
7
7
5
7
.
1
4
3
3
.
3
3
0
.
0
3
9
7
7
7
6
.
1
9
3
3
.
3
3
4
0
.
0
7
6
5
4
6
1
.
9
0
5
0
.
0
0
0
.
0
4
1
6
3
7
1
.
4
3
5
0
.
0
0
5
0
.
0
8
3
2
1
6
1
.
9
0
3
3
.
3
3
0
.
0
3
8
9
2
7
6
.
1
9
5
0
.
0
0
6
0
.
0
7
8
6
6
6
6
.
6
7
5
0
.
0
0
0
.
0
4
0
1
1
7
1
.
4
3
3
3
.
3
3
D
F
F
N
N
+
P
S
O
Te
st
F
i
n
a
l
M
S
E
Tr
a
i
n
i
n
g
A
c
c
u
r
a
c
y
(
%)
Te
st
A
c
c
u
r
a
c
y
(
%)
1
0
.
0
3
4
7
8
7
6
.
1
9
3
3
.
3
3
2
0
.
0
3
1
6
6
8
0
.
9
5
5
0
.
0
0
3
0
.
0
2
9
8
4
8
0
.
9
5
3
3
.
3
3
4
0
.
0
3
3
1
2
7
6
.
1
9
5
0
.
0
0
5
0
.
0
3
0
7
7
8
0
.
9
5
5
0
.
0
0
6
0
.
0
3
2
9
1
7
6
.
1
9
3
3
.
3
3
Te
st
D
F
F
N
N
+
G
W
O
D
F
F
N
N
+
G
A
+
G
W
O
F
i
n
a
l
M
S
E
Tr
a
i
n
i
n
g
A
c
c
u
r
a
c
y
(
%)
Te
st
A
c
c
u
r
a
c
y
(
%)
F
i
n
a
l
M
S
E
Tr
a
i
n
i
n
g
A
c
c
u
r
a
c
y
(
%)
Te
st
A
c
c
u
r
a
c
y
(
%)
1
0
.
0
2
4
1
1
8
0
.
9
5
5
0
.
0
0
0
.
0
1
7
2
1
8
0
.
9
5
5
0
.
0
0
2
0
.
0
2
1
7
7
8
5
.
7
1
5
0
.
0
0
0
.
0
1
9
3
4
7
6
.
1
9
3
3
.
3
3
3
0
.
0
1
9
8
8
8
5
.
7
1
3
3
.
3
3
0
.
0
1
6
8
8
8
5
.
7
1
5
0
.
0
0
4
0
.
0
2
2
3
1
8
0
.
9
5
5
0
.
0
0
0
.
0
1
8
7
7
8
0
.
9
5
5
0
.
0
0
5
0
.
0
1
8
9
4
8
5
.
7
1
5
0
.
0
0
0
.
0
1
5
9
9
8
5
.
7
1
3
3
.
3
3
6
0
.
0
2
0
6
3
8
0
.
9
5
3
3
.
3
3
0
.
0
1
7
4
3
8
0
.
9
5
5
0
.
0
0
I
n
g
en
e
r
al,
th
e
ap
p
licatio
n
o
f
s
in
g
le
-
s
tag
e
m
etah
eu
r
is
tic
o
p
tim
izatio
n
im
p
r
o
v
ed
th
ese
li
m
itatio
n
s
.
T
h
e
DFFNN+GA
m
o
d
el
d
em
o
n
s
tr
ated
a
r
a
p
id
MSE
r
ed
u
ctio
n
d
u
r
in
g
t
h
e
ea
r
ly
iter
atio
n
s
,
h
ig
h
lig
h
tin
g
GA’
s
s
tr
o
n
g
ca
p
ab
ilit
y
f
o
r
g
lo
b
al
e
x
p
lo
r
atio
n
in
th
e
s
o
lu
tio
n
s
p
ac
e.
Ho
wev
er
,
d
esp
ite
its
ef
f
ec
tiv
en
ess
in
r
ed
u
cin
g
th
e
in
itial
er
r
o
r
,
th
e
f
in
al
p
e
r
f
o
r
m
an
ce
s
till
s
h
o
wed
n
o
ticea
b
le
v
ar
iatio
n
ac
r
o
s
s
tr
ials
.
T
h
i
s
r
ef
lects
th
e
h
ig
h
l
y
s
to
ch
asti
c
n
atu
r
e
o
f
GA,
wh
e
r
e
ag
g
r
ess
iv
e
ex
p
lo
r
atio
n
is
n
o
t
alwa
y
s
ac
co
m
p
an
ied
b
y
a
s
tab
le
ex
p
lo
itatio
n
p
h
ase
th
at
co
n
s
is
ten
tly
r
ef
in
es
th
e
s
o
lu
tio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.