I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
15
,
No
.
4
,
A
u
g
u
s
t
20
26
,
p
p
.
3
2
2
8
~
3
2
3
9
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
4
.
p
p
3
2
2
8
-
3
2
3
9
3228
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
H
y
brid d
eep l
ea
rning
mo
del f
o
r
en
ha
nced sho
rt
-
ter
m g
o
ld price
foreca
sting
H
o
a
ng
H
a
Ng
uy
en
1
,
M
inh
D
uc
Ng
uy
en
2
,
Cuo
ng
H
.
Ng
uy
en
-
Dinh
3
1
F
a
c
u
l
t
y
o
f
I
n
f
o
r
m
a
t
i
o
n
T
e
c
h
n
o
l
o
g
y
,
U
n
i
v
e
r
si
t
y
o
f
S
c
i
e
n
c
e
s
,
H
u
e
U
n
i
v
e
r
s
i
t
y
,
H
u
e
,
V
i
e
t
n
a
m
2
D
e
p
a
r
t
me
n
t
o
f
Ec
o
n
o
m
i
c
I
n
f
o
r
mat
i
o
n
S
y
st
e
ms,
U
n
i
v
e
r
s
i
t
y
o
f
E
c
o
n
o
mi
c
s,
H
u
e
U
n
i
v
e
r
s
i
t
y
,
H
u
e
,
V
i
e
t
n
a
m
3
D
e
p
a
r
t
me
n
t
o
f
D
a
t
a
S
c
i
e
n
c
e
,
U
n
i
v
e
r
s
i
t
y
o
f
F
i
n
a
n
c
e
-
M
a
r
k
e
t
i
n
g
,
H
o
C
h
i
M
i
n
h
,
V
i
e
t
n
a
m
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
l
2
,
2
0
2
5
R
ev
is
ed
J
u
l
13
,
2
0
2
6
Acc
ep
ted
J
u
l
21
,
2
0
2
6
Ac
c
u
ra
te
sh
o
rt
-
term
g
o
l
d
p
r
ice
fo
re
c
a
stin
g
is
c
ru
c
ial
f
o
r
t
h
e
fin
a
n
c
ial
d
e
c
isio
n
-
m
a
k
i
n
g
.
Th
is
p
a
p
e
r
in
tr
o
d
u
c
e
s
a
sh
o
rt
-
term
g
o
ld
p
re
d
ictio
n
n
e
tw
o
rk
(
S
TG
P
-
Ne
t
)
,
a
n
o
v
e
l
h
y
b
ri
d
d
e
e
p
lea
rn
in
g
m
o
d
e
l
d
e
sig
n
e
d
t
o
e
n
h
a
n
c
e
p
re
d
ictio
n
a
c
c
u
ra
c
y
b
y
i
n
teg
ra
ti
n
g
o
n
e
-
d
ime
n
si
o
n
a
l
c
o
n
v
o
lu
t
io
n
a
l
n
e
u
ra
l
n
e
two
rk
(
1D
-
CNN
)
a
n
d
l
o
n
g
sh
o
rt
-
term
m
e
m
o
ry
(
LS
T
M
)
.
S
TG
P
-
Ne
t
lev
e
ra
g
e
s
th
e
1
D
-
CNN
'
s
a
b
il
it
y
to
e
x
trac
t
lo
c
a
l
tem
p
o
ra
l
fe
a
tu
re
s
a
n
d
th
e
LS
TM
c
a
p
a
c
it
y
t
o
m
o
d
e
l
l
o
n
g
-
r
a
n
g
e
d
e
p
e
n
d
e
n
c
ies
wit
h
i
n
g
o
l
d
p
rice
ti
m
e
se
ries
.
Va
rio
u
s
slid
i
n
g
wi
n
d
o
w
c
o
n
fi
g
u
ra
ti
o
n
s
a
re
e
m
p
lo
y
e
d
t
o
g
e
n
e
ra
te
in
p
u
t
se
q
u
e
n
c
e
s fo
r
m
u
lt
i
-
ste
p
a
h
e
a
d
p
re
d
ictio
n
.
C
o
m
p
re
h
e
n
si
v
e
e
x
p
e
rime
n
ts
we
re
c
o
n
d
u
c
ted
c
o
m
p
a
rin
g
S
TGP
-
Ne
t
a
g
a
in
st
1
D
-
CNN
+
re
c
u
rre
n
t
n
e
u
ra
l
n
e
two
rk
(
RNN
)
a
n
d
1
D
-
CNN
+
b
i
d
irec
ti
o
n
a
l
l
o
n
g
s
h
o
rt
-
term
m
e
m
o
ry
(
BiL
S
TM
)
b
a
se
li
n
e
m
o
d
e
ls
a
c
ro
ss
th
re
e
c
o
n
fig
u
ra
ti
o
n
s
u
sin
g
m
e
tri
c
s
li
k
e
m
e
a
n
a
b
so
lu
te
e
rro
r
(
M
AE
)
,
ro
o
t
m
e
a
n
sq
u
a
re
e
rro
r
(
R
M
S
E
)
,
a
n
d
d
e
term
in
a
ti
o
n
(
R²
)
.
Th
e
re
su
l
ts
d
e
m
o
n
stra
ted
th
a
t
S
TG
P
-
Ne
t
c
o
n
siste
n
tl
y
p
ro
v
id
e
d
b
e
tt
e
r
p
e
rfo
rm
a
n
c
e
a
n
d
ro
b
u
st
n
e
ss
,
p
ro
v
in
g
m
o
re
e
ff
e
c
ti
v
e
fo
r
sh
o
rt
-
term
g
o
ld
p
rice
f
o
re
c
a
stin
g
th
a
n
t
h
e
a
lt
e
rn
a
ti
v
e
h
y
b
ri
d
m
o
d
e
ls t
e
ste
d
.
K
ey
w
o
r
d
s
:
1D
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etw
o
r
k
Go
ld
p
r
ice
p
r
ed
ictio
n
Hy
b
r
id
d
ee
p
lear
n
i
n
g
L
o
n
g
s
h
o
r
t
-
ter
m
m
e
m
o
r
y
T
im
e
s
er
ies f
o
r
ec
asti
n
g
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
C
u
o
n
g
H.
Ng
u
y
en
-
Din
h
Dep
ar
tm
en
t o
f
Data
Scien
ce
,
Un
iv
er
s
ity
o
f
Fin
an
ce
–
Ma
r
k
etin
g
7
7
8
Ng
u
y
en
Kiem
St.,
D
u
c
N
h
u
an
W
ar
d
,
H
o
C
h
i M
in
h
C
ity
,
Vietn
am
E
m
ail: c
u
o
n
g
n
d
h
@
u
f
m
.
e
d
u
.
v
n
1.
I
NT
RO
D
UCT
I
O
N
R
ec
en
t
ad
v
an
ce
m
en
ts
in
d
ee
p
lear
n
in
g
h
av
e
h
i
g
h
lig
h
ted
th
e
s
ig
n
if
ican
t
ad
v
an
tag
es
o
f
h
y
b
r
id
m
o
d
els
th
at
co
m
b
in
e
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN)
ar
ch
itec
tu
r
es
f
o
r
s
p
atial
f
ea
tu
r
e
ex
tr
ac
tio
n
with
r
ec
u
r
r
en
t
n
eu
r
al
n
etwo
r
k
(
R
NN)
ar
c
h
itectu
r
es
f
o
r
tem
p
o
r
al
s
eq
u
en
c
e
m
o
d
elin
g
;
th
ese
s
y
n
er
g
is
tic
ap
p
r
o
ac
h
es
h
av
e
en
h
an
ce
d
p
er
f
o
r
m
a
n
ce
ac
r
o
s
s
d
iv
er
s
e
r
esear
ch
d
o
m
ai
n
s
r
an
g
in
g
f
r
o
m
s
en
tim
en
t
a
n
aly
s
is
[
1
]
an
d
o
b
ject
d
etec
tio
n
in
v
id
eo
[
2
]
to
a
u
d
i
o
p
r
o
ce
s
s
in
g
[
3
]
an
d
m
ed
ical
im
ag
e
an
aly
s
is
[
4
]
.
T
h
is
tr
en
d
h
as
ex
ten
d
e
d
in
to
th
e
f
in
a
n
cial
s
ec
to
r
,
wh
er
e
a
cc
u
r
ately
p
r
ed
ictin
g
tim
e
s
er
ies
d
ata
is
cr
u
cial
[
5
]
.
Nu
m
er
o
u
s
s
tu
d
ies
h
a
v
e
ex
p
lo
r
ed
h
y
b
r
id
m
o
d
els
co
m
b
in
in
g
elem
en
ts
f
r
o
m
th
e
C
NN
an
d
R
NN
f
am
ilies
f
o
r
f
o
r
ec
asti
n
g
v
o
latile
f
in
an
cial
m
ar
k
ets,
in
clu
d
in
g
cr
y
p
to
cu
r
r
en
c
y
p
r
ice
p
r
e
d
ictio
n
[
6
]
,
s
to
ck
m
ar
k
et
an
aly
s
is
[
7
]
,
an
d
co
m
m
o
d
ity
f
o
r
ec
asti
n
g
,
s
u
ch
as
g
o
ld
p
r
ice
p
r
ed
ictio
n
[
8
]
.
W
ith
in
th
e
C
NN
f
am
ily
,
o
n
e
-
d
im
en
s
io
n
al
c
o
n
v
o
lu
ti
o
n
al
n
e
u
r
al
n
etwo
r
k
s
(
1
D
-
C
NNs)
p
o
s
s
ess
a
d
is
tin
ct
ad
v
an
tag
e
f
o
r
tim
e
s
er
ies
an
aly
s
is
,
as
th
eir
k
er
n
els
o
p
er
ate
alo
n
g
th
e
tem
p
o
r
al
d
im
e
n
s
io
n
,
m
ak
in
g
t
h
em
h
ig
h
l
y
ef
f
ec
tiv
e
at
ex
t
r
ac
tin
g
lo
ca
l
p
atter
n
s
an
d
r
elev
a
n
t
f
ea
tu
r
es
d
ir
ec
tly
f
r
o
m
s
eq
u
en
tial
d
ata
[
9
]
.
R
ec
o
g
n
izin
g
th
e
p
o
ten
tial
s
y
n
er
g
y
b
etwe
en
1
D
-
C
NNs
'
f
ea
tu
r
e
ex
tr
ac
tio
n
p
o
wer
an
d
th
e
s
eq
u
en
ce
-
h
an
d
lin
g
ca
p
ab
ilit
ies
o
f
d
if
f
er
en
t
R
NN
v
ar
ian
ts
,
th
is
s
tu
d
y
ex
p
lo
r
es
an
d
ev
alu
ates
h
y
b
r
id
m
o
d
els
co
m
b
in
i
n
g
1
D
-
C
NN
with
m
em
b
er
s
o
f
th
e
R
NN
f
a
m
ily
—
s
p
ec
if
ically
s
tan
d
ar
d
R
NN,
lo
n
g
s
h
o
r
t
-
ter
m
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Hyb
r
id
d
ee
p
lea
r
n
in
g
mo
d
el
f
o
r
en
h
a
n
ce
d
s
h
o
r
t
-
term g
o
ld
p
r
ice
…
(
Ho
a
n
g
Ha
N
g
u
ye
n
)
3229
m
em
o
r
y
(
L
STM
)
,
a
n
d
b
id
ir
ec
tio
n
al
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
B
iLST
M)
—
to
id
en
tify
a
r
o
b
u
s
t
an
d
ef
f
ec
tiv
e
ca
n
d
id
ate
m
o
d
el
f
o
r
th
e
ch
alle
n
g
in
g
task
o
f
s
h
o
r
t
-
ter
m
g
o
ld
p
r
ice
f
o
r
ec
asti
n
g
.
Go
ld
h
o
l
d
s
s
ig
n
if
ican
t
im
p
o
r
t
an
ce
in
g
lo
b
al
f
in
an
cial
la
n
d
s
ca
p
e,
wid
ely
r
ec
o
g
n
ized
as
a
s
af
e
-
h
av
en
ass
et
an
d
a
cr
u
cial
to
o
l
f
o
r
h
ed
g
in
g
ag
ain
s
t
in
f
latio
n
,
cu
r
r
en
cy
f
lu
ctu
atio
n
s
,
an
d
g
eo
p
o
liti
ca
l
u
n
ce
r
tain
ty
.
C
o
n
s
eq
u
en
tly
,
th
e
ac
cu
r
ate
p
r
ed
ictio
n
o
f
its
p
r
ice
m
o
v
em
en
ts
,
p
ar
ticu
lar
ly
in
th
e
s
h
o
r
t
ter
m
,
is
o
f
co
n
s
id
er
ab
le
v
alu
e
to
i
n
v
esto
r
s
,
tr
ad
er
s
,
p
o
r
tf
o
lio
m
an
a
g
er
s
,
an
d
p
o
licy
m
ak
er
s
f
o
r
m
ak
in
g
in
f
o
r
m
ed
d
ec
is
io
n
s
,
m
an
ag
in
g
r
is
k
,
an
d
ca
p
italizin
g
o
n
m
ar
k
et
o
p
p
o
r
tu
n
ities
.
Ad
d
r
ess
in
g
th
is
n
ee
d
,
th
is
p
ap
er
in
tr
o
d
u
ce
s
th
e
s
h
o
r
t
-
ter
m
g
o
ld
p
r
ed
ictio
n
n
etwo
r
k
(
ST
GP
-
Net)
,
a
n
o
v
el
h
y
b
r
id
d
e
ep
lear
n
in
g
m
o
d
el
s
p
ec
if
ically
d
esig
n
ed
f
o
r
e
n
h
a
n
ce
d
s
h
o
r
t
-
ter
m
g
o
ld
p
r
ice
f
o
r
ec
asti
n
g
.
T
h
e
m
ajo
r
co
n
tr
ib
u
tio
n
s
o
f
th
is
s
tu
d
y
ar
e
th
r
ee
f
o
ld
:
i
)
p
r
o
p
o
s
e
th
e
n
o
v
el
STGP
-
Net
ar
ch
itectu
r
e,
in
teg
r
atin
g
a
1
D
-
C
NN
with
L
STM
to
ef
f
ec
tiv
ely
ca
p
tu
r
e
b
o
th
lo
ca
l
tem
p
o
r
al
f
e
atu
r
es
an
d
lo
n
g
-
r
an
g
e
d
e
p
en
d
en
cies
in
g
o
ld
p
r
ice
d
ata
;
ii
)
s
y
s
tem
atica
lly
ap
p
ly
d
if
f
er
en
t
s
lid
in
g
win
d
o
w
s
ca
les
(
co
n
f
ig
u
r
atio
n
s
)
to
g
e
n
er
at
e
in
p
u
t
d
ata,
e
n
ab
lin
g
th
e
m
o
d
el
to
lear
n
f
r
o
m
v
ar
y
in
g
h
is
to
r
ical
s
eq
u
en
ce
len
g
th
s
f
o
r
m
u
lti
-
s
tep
ah
ea
d
p
r
ed
ictio
n
;
an
d
iii
)
co
n
d
u
ct
co
m
p
r
e
h
en
s
iv
e
ex
p
er
im
en
ts
c
o
m
p
ar
i
n
g
STG
P
-
Net
ag
ain
s
t
two
r
elev
an
t
b
aselin
e
h
y
b
r
id
m
o
d
els
(
1
D
-
C
NN
+
R
NN
an
d
1D
-
C
NN
+
B
iL
STM
)
,
d
em
o
n
s
tr
atin
g
th
e
s
u
p
er
io
r
p
e
r
f
o
r
m
a
n
ce
an
d
r
o
b
u
s
tn
ess
o
f
p
r
o
p
o
s
ed
m
o
d
el
th
r
o
u
g
h
r
ig
o
r
o
u
s
ev
alu
atio
n
m
etr
ics.
T
h
e
r
est
o
f
t
h
is
p
ap
er
is
o
r
g
a
n
ized
as
f
o
llo
ws.
Sectio
n
2
e
lab
o
r
ates
o
n
t
h
e
s
tate
-
of
-
th
e
-
ar
t
s
tu
d
ies
r
elev
an
t
to
h
y
b
r
id
d
ee
p
lear
n
i
n
g
m
o
d
els
f
o
r
f
in
an
cial
tim
e
s
er
ies
f
o
r
ec
asti
n
g
,
p
r
o
v
id
in
g
co
n
tex
t
f
o
r
t
h
is
wo
r
k
.
Sectio
n
3
in
tr
o
d
u
ce
s
th
e
p
r
o
p
o
s
ed
STGP
-
Net
ar
ch
itectu
r
e
in
d
etail,
elab
o
r
atin
g
o
n
its
co
n
s
titu
en
t
1
D
-
C
NN
an
d
L
STM
co
m
p
o
n
e
n
ts
an
d
ex
p
lain
in
g
th
e
m
ec
h
a
n
is
m
b
y
wh
ich
it
p
r
o
ce
s
s
es
d
ata
f
o
r
p
r
ed
ictio
n
.
Sectio
n
4
p
r
esen
ts
th
e
ex
p
er
im
en
tal
s
etu
p
,
in
clu
d
in
g
d
ata
p
r
ep
ar
atio
n
,
co
n
f
ig
u
r
atio
n
s
,
b
aselin
e
m
o
d
els,
an
d
ev
alu
atio
n
m
etr
ics,
f
o
llo
wed
b
y
a
p
r
esen
t
atio
n
an
d
d
is
cu
s
s
io
n
o
f
th
e
c
o
m
p
ar
ativ
e
r
esu
lts
.
Fin
ally
,
s
ec
tio
n
5
co
n
clu
d
es th
e
p
ap
er
b
y
s
u
m
m
ar
izin
g
th
e
k
ey
f
in
d
in
g
s
an
d
s
u
g
g
esti
n
g
p
o
ten
tial d
ir
ec
tio
n
s
f
o
r
f
u
tu
r
e
r
esear
ch
.
2.
RE
L
AT
E
D
WO
RK
T
h
is
s
ec
tio
n
p
r
o
v
id
es
a
r
ev
iew
o
f
th
e
r
elev
an
t
liter
atu
r
e
co
n
ce
r
n
in
g
m
eth
o
d
o
lo
g
ies
em
p
lo
y
ed
f
o
r
f
o
r
ec
asti
n
g
g
o
ld
p
r
ices,
a
ta
s
k
th
at
h
as
attr
ac
ted
s
ig
n
if
i
ca
n
t
r
esear
ch
in
ter
est
d
u
e
t
o
g
o
ld
'
s
ec
o
n
o
m
ic
im
p
o
r
tan
ce
.
T
h
e
ev
o
lu
tio
n
o
f
p
r
ed
ictiv
e
m
o
d
els
ap
p
lied
to
th
is
ch
allen
g
e
is
ex
am
in
ed
,
co
v
er
in
g
tr
a
d
itio
n
al
s
tatis
t
ical
tim
e
s
er
ies
tech
n
iq
u
es,
co
n
v
e
n
tio
n
al
m
ac
h
in
e
lear
n
in
g
al
g
o
r
ith
m
s
,
an
d
m
o
r
e
r
e
ce
n
t
ad
v
a
n
ce
m
en
ts
in
d
ee
p
lear
n
in
g
,
in
clu
d
in
g
th
e
d
ev
elo
p
m
en
t o
f
h
y
b
r
i
d
ar
c
h
itectu
r
es.
T
h
is
o
v
e
r
v
iew
aim
s
t
o
s
itu
ate
th
e
p
r
esen
t
s
tu
d
y
with
in
th
e
lan
d
s
ca
p
e
o
f
ex
is
tin
g
r
esear
ch
an
d
h
ig
h
li
g
h
t
th
e
co
n
tex
t
f
o
r
ex
p
l
o
r
in
g
n
o
v
el
h
y
b
r
id
d
ee
p
lear
n
in
g
ap
p
r
o
ac
h
es f
o
r
e
n
h
an
ce
d
s
h
o
r
t
-
ter
m
g
o
ld
p
r
ice
p
r
ed
ictio
n
.
His
to
r
ically
,
g
o
ld
p
r
ice
f
o
r
ec
asti
n
g
was
p
r
im
ar
ily
ad
d
r
ess
ed
u
s
in
g
tr
ad
itio
n
al
s
tatis
tical
t
im
e
s
er
ies
m
o
d
els.
Fo
r
ex
am
p
le,
Ma
k
ala
an
d
L
i
[
1
0
]
co
m
p
ar
e
d
th
e
p
er
f
o
r
m
an
ce
o
f
au
to
r
e
g
r
ess
iv
e
in
teg
r
ated
m
o
v
in
g
av
er
ag
e
(
AR
I
MA
)
an
d
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
(
SVM
)
m
o
d
els
f
o
r
p
r
ed
ictin
g
d
aily
g
o
ld
p
r
ices
u
s
in
g
d
ata
s
p
an
n
in
g
f
r
o
m
1
9
7
9
to
2
0
1
9
.
T
h
eir
r
esu
lts
in
d
icate
d
th
at
th
e
SVM
m
o
d
el
s
ig
n
if
ican
tly
o
u
tp
er
f
o
r
m
e
d
th
e
AR
I
MA
m
o
d
el.
B
ased
o
n
s
u
b
s
tan
tially
lo
wer
r
o
o
t
m
ea
n
s
q
u
ar
e
er
r
o
r
(
R
MSE
)
an
d
m
ea
n
ab
s
o
lu
te
p
er
ce
n
tag
e
er
r
o
r
(
MA
PE
)
v
alu
es,
t
h
e
au
t
h
o
r
s
co
n
clu
d
ed
t
h
at
SVM
is
a
m
o
r
e
ac
c
u
r
ate
an
d
s
u
itab
le
m
eth
o
d
f
o
r
g
o
ld
p
r
ice
p
r
ed
ictio
n
.
Su
ch
f
i
n
d
in
g
s
s
u
g
g
ested
th
at
ev
en
co
n
v
en
tio
n
al
m
ac
h
in
e
lear
n
in
g
m
o
d
els
lik
e
SVM
co
u
ld
o
f
f
er
im
p
r
o
v
e
d
p
r
e
d
ictiv
e
ca
p
ab
ilit
i
es
o
v
er
tr
ad
itio
n
al
s
tatis
tical
ti
m
e
s
er
ies
m
eth
o
d
s
lik
e
AR
I
MA
f
o
r
h
an
d
lin
g
th
e
co
m
p
lex
ities
in
h
er
en
t i
n
g
o
l
d
p
r
ice
d
ata.
As
a
m
ac
h
in
e
lear
n
in
g
ap
p
r
o
ac
h
to
ad
d
r
ess
th
e
lim
itatio
n
s
o
f
g
r
ad
ien
t
d
escen
t
in
n
e
u
r
a
l
n
etwo
r
k
o
p
tim
izatio
n
,
Do
u
s
h
et
a
l.
[
1
1
]
p
r
o
p
o
s
ed
an
ar
ch
iv
e
-
b
a
s
ed
Har
r
is
h
awk
s
o
p
tim
izer
(
AHHO
)
to
tu
n
e
m
u
lti
-
lay
er
p
er
ce
p
tr
o
n
(
ML
P)
p
ar
am
eter
s
f
o
r
g
o
l
d
p
r
ice
f
o
r
ec
asti
n
g
.
T
h
e
r
esear
c
h
er
s
e
m
p
lo
y
ed
Pear
s
o
n
’
s
co
r
r
elatio
n
a
n
d
a
ca
teg
o
r
ized
co
r
r
elatio
n
f
o
r
f
ea
tu
r
e
s
elec
tio
n
an
d
b
en
ch
m
ar
k
ed
th
e
m
o
d
el
ag
ain
s
t
eig
h
t
alter
n
ativ
e
m
ac
h
in
e
lear
n
i
n
g
an
d
s
war
m
in
tellig
en
ce
alg
o
r
ith
m
s
.
T
h
eir
f
in
d
in
g
s
d
em
o
n
s
tr
ated
th
at
th
e
AHHO
-
n
eu
r
al
n
etwo
r
k
m
o
d
e
l
p
r
o
d
u
ce
d
h
ig
h
er
ac
c
u
r
ac
y
an
d
lo
wer
er
r
o
r
m
etr
ics
th
a
n
th
e
c
o
m
p
a
r
ativ
e
m
o
d
els,
s
u
g
g
esti
n
g
th
at
th
e
in
teg
r
atio
n
o
f
an
ar
c
h
iv
e
ef
f
ec
tiv
ely
im
p
r
o
v
es
p
er
f
o
r
m
an
ce
in
co
m
p
lex
f
o
r
ec
asti
n
g
task
s
.
T
o
f
u
r
t
h
er
th
e
d
ev
elo
p
m
en
t
o
f
m
ac
h
i
n
e
lear
n
in
g
tr
en
d
s
in
f
o
r
ec
asti
n
g
f
i
n
an
cial
p
r
ice
d
ata,
r
ec
en
t
liter
atu
r
e
h
as
s
h
if
te
d
t
o
war
d
h
y
b
r
id
f
r
am
ewo
r
k
s
d
e
s
ig
n
ed
to
ca
p
tu
r
e
th
e
v
o
latilit
y
an
d
m
u
lti
-
s
ca
le
d
ep
en
d
e
n
cies
in
h
er
en
t
i
n
th
e
s
e
s
er
ies.
W
h
ile
Gh
o
s
h
an
d
J
an
a
[
1
2
]
f
o
cu
s
ed
o
n
t
h
e
ex
tr
em
e
v
o
latilit
y
o
f
h
ig
h
-
f
r
eq
u
e
n
cy
in
tr
ad
ay
m
o
v
em
en
ts
b
y
u
tili
zin
g
m
ax
im
a
l
o
v
er
la
p
d
is
cr
ete
wav
elet
t
r
an
s
f
o
r
m
atio
n
an
d
B
ay
esian
s
tr
u
ctu
r
al
m
o
d
els
to
m
ai
n
tain
p
r
ed
ictiv
e
r
esil
ien
ce
d
u
r
in
g
g
lo
b
al
e
co
n
o
m
ic
s
h
o
ck
s
,
So
n
g
an
d
C
h
en
[
1
3
]
ad
d
r
ess
ed
th
e
b
r
o
ad
er
tem
p
o
r
al
f
ea
tu
r
es
o
f
th
e
s
h
ip
p
in
g
m
a
r
k
et
b
y
em
p
lo
y
in
g
v
ar
iatio
n
al
m
o
d
e
d
ec
o
m
p
o
s
itio
n
(
VM
D)
with
lig
h
t
g
r
ad
ien
t
b
o
o
s
tin
g
m
ac
h
in
e
(
L
ig
h
tGB
M)
to
ex
tr
ac
t
m
u
lti
-
s
ca
le
in
f
o
r
m
atio
n
f
r
o
m
lo
wer
-
f
r
eq
u
en
cy
tim
e
s
er
ies.
Dis
t
in
g
u
is
h
in
g
its
elf
f
r
o
m
th
ese
d
ec
o
m
p
o
s
itio
n
-
ce
n
tr
ic
m
eth
o
d
s
,
th
e
wo
r
k
o
f
B
h
am
b
u
et
a
l.
[
1
4
]
i
n
tr
o
d
u
ce
d
th
e
r
e
cu
r
r
en
t
e
n
s
em
b
le
d
ee
p
r
an
d
o
m
v
ec
to
r
f
u
n
ctio
n
al
lin
k
(
R
ed
R
VFL)
n
etwo
r
k
,
w
h
ich
lev
er
ag
es
s
tab
le,
f
ix
ed
-
weig
h
t
r
ec
u
r
r
e
n
t
lay
er
s
an
d
d
ee
p
r
ep
r
esen
tatio
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t
20
26
:
3
2
2
8
-
3
2
3
9
3230
lear
n
in
g
to
ac
h
iev
e
s
u
p
er
i
o
r
ac
cu
r
ac
y
ac
r
o
s
s
d
iv
er
s
e
f
in
an
cial
d
atasets
with
o
u
t
r
eq
u
ir
in
g
th
e
s
p
ec
if
ic
s
ig
n
al
-
p
r
o
ce
s
s
in
g
s
tag
es u
s
ed
in
th
e
af
o
r
em
en
tio
n
e
d
s
tu
d
ies.
C
o
h
en
an
d
Aich
e
[
1
5
]
ex
p
l
o
r
ed
ad
v
a
n
ce
d
m
ac
h
in
e
lear
n
in
g
tech
n
iq
u
es,
s
p
ec
if
ically
co
m
p
ar
in
g
r
an
d
o
m
f
o
r
est
(
R
F)
,
g
r
ad
ien
t
b
o
o
s
ted
r
eg
r
ess
io
n
tr
ee
s
(
GB
R
T
)
,
an
d
ex
tr
em
e
g
r
a
d
ien
t
b
o
o
s
tin
g
(
XGBo
o
s
t)
f
o
r
f
o
r
ec
asti
n
g
g
o
l
d
p
r
ices.
T
h
e
m
o
d
els
u
tili
ze
d
a
co
m
p
r
e
h
en
s
iv
e
s
et
o
f
in
p
u
t
f
ea
tu
r
es
in
clu
d
in
g
lag
g
e
d
g
lo
b
al
s
to
ck
in
d
ices,
v
o
latilit
y
in
d
e
x
(
VI
X
)
,
co
m
m
o
d
ity
f
u
t
u
r
es,
an
d
b
o
n
d
y
ield
s
f
r
o
m
s
ev
e
r
al
m
a
jo
r
ec
o
n
o
m
ies.
T
h
e
r
esear
ch
id
e
n
tifie
d
th
at
ce
r
tai
n
lag
g
e
d
s
to
c
k
in
d
ices
(
ASX
an
d
S&
P5
0
0
)
,
US/J
ap
an
b
o
n
d
y
ield
s
,
g
as/s
ilv
er
p
r
ices,
an
d
th
e
VI
X
wer
e
p
ar
ticu
lar
ly
in
f
lu
en
tial
p
r
ed
icto
r
s
,
co
n
f
ir
m
in
g
th
e
im
p
ac
t
o
f
ec
o
n
o
m
ic
u
n
ce
r
tain
t
y
.
Ultim
ately
,
th
e
s
tu
d
y
co
n
clu
d
ed
th
at
en
s
em
b
le
m
o
d
els
lik
e
GB
R
T
an
d
XG
B
o
o
s
t
p
r
o
v
ed
v
alu
ab
le
f
o
r
m
ak
in
g
in
f
o
r
m
e
d
g
o
ld
in
v
estme
n
t
d
ec
is
io
n
s
b
ased
o
n
t
h
ese
d
iv
er
s
e
f
in
an
cial
in
d
icato
r
s
.
A
co
m
p
r
eh
en
s
iv
e
o
v
e
r
v
iew
s
u
m
m
ar
izin
g
th
e
ap
p
licatio
n
o
f
v
ar
io
u
s
m
ac
h
in
e
lear
n
in
g
tech
n
iq
u
es
to
g
o
l
d
p
r
ice
p
r
ed
ictio
n
o
v
er
th
e
p
r
ec
ed
in
g
d
ec
ad
e
was p
r
o
v
id
e
d
[
1
6
]
.
Mo
r
e
r
ec
en
tly
,
th
e
f
ield
h
as
witn
ess
ed
a
s
ig
n
if
ican
t
s
h
if
t
to
war
d
s
d
ee
p
lear
n
in
g
m
eth
o
d
o
lo
g
ies,
lev
er
ag
in
g
co
m
p
le
x
n
e
u
r
al
n
e
two
r
k
ar
ch
itectu
r
es
to
p
o
ten
tially
ca
p
tu
r
e
m
o
r
e
i
n
tr
icate
te
m
p
o
r
al
p
atter
n
s
in
g
o
ld
p
r
ice
d
ata.
Fo
r
ex
am
p
le
,
C
h
en
an
d
Hu
a
n
g
[
1
7
]
co
m
p
ar
ed
C
NN
an
d
L
STM
m
o
d
e
ls
,
o
p
tim
ized
u
s
in
g
B
ay
esian
o
p
tim
izatio
n
,
to
f
o
r
ec
ast
S&
P
5
0
0
m
o
v
em
en
t
s
b
ased
o
n
v
ar
io
u
s
in
p
u
ts
in
clu
d
in
g
f
in
a
n
cial
in
d
icato
r
s
as
well
as
g
o
ld
a
n
d
cr
u
d
e
o
il
p
r
ice
an
d
v
o
latilit
y
in
d
ices.
T
h
e
m
o
d
els
wer
e
tr
a
in
ed
o
n
d
ata
f
r
o
m
2010
-
2
0
1
8
an
d
test
ed
o
n
2
0
1
9
d
ata.
E
x
p
er
im
en
tal
r
esu
lts
,
ev
alu
ated
b
ased
o
n
class
if
icatio
n
ac
cu
r
ac
y
an
d
b
ac
k
-
test
ed
r
etu
r
n
o
n
in
v
estme
n
t
(
R
OI
)
,
s
h
o
wed
th
at
th
e
p
r
o
p
o
s
ed
C
NN
m
o
d
el
with
eig
h
t
f
ea
tu
r
es
(
C
NN8
)
ac
h
iev
ed
a
h
ig
h
er
R
OI
(
1
3
.
2
3
%)
th
an
th
e
eq
u
iv
alen
t
L
STM
m
o
d
el
(
1
1
.
0
6
%).
T
h
e
au
th
o
r
s
also
n
o
ted
th
at
in
clu
d
in
g
g
o
ld
an
d
o
il
-
r
elate
d
f
ea
tu
r
es
p
r
o
v
e
d
p
ar
ticu
lar
l
y
b
en
ef
icial
f
o
r
p
r
ed
ictio
n
s
r
elate
d
to
ce
r
tain
in
d
u
s
tr
ies lik
e
s
em
ico
n
d
u
cto
r
s
an
d
p
etr
o
leu
m
.
B
ey
o
n
d
c
o
m
p
ar
i
n
g
estab
lis
h
ed
m
o
d
els
lik
e
C
NN
an
d
L
STM
,
r
esear
ch
also
ex
p
lo
r
ed
n
o
v
el
s
tan
d
alo
n
e
d
ee
p
lear
n
i
n
g
ar
c
h
itectu
r
es
d
esig
n
ed
to
ca
p
tu
r
e
co
m
p
lex
m
ar
k
et
d
y
n
am
ics
d
i
r
ec
tly
.
Fo
r
in
s
tan
ce
,
Hu
an
g
et
a
l.
[
1
8
]
p
r
o
p
o
s
ed
a
n
o
v
el
m
u
ltil
ev
el
g
r
ap
h
atten
ti
o
n
n
etwo
r
k
(
ML
-
GAT
)
to
a
d
d
r
ess
lim
itatio
n
s
in
ex
is
tin
g
g
r
ap
h
-
b
ased
s
to
ck
p
r
ed
ictio
n
m
et
h
o
d
s
,
p
a
r
ticu
lar
ly
r
eg
a
r
d
in
g
t
h
e
in
co
r
p
o
r
atio
n
o
f
d
iv
er
s
e
in
f
o
r
m
atio
n
an
d
s
to
ck
r
elatio
n
s
h
ip
s
d
er
iv
ed
f
r
o
m
k
n
o
wled
g
e
g
r
a
p
h
s
lik
e
W
ik
id
ata
.
A
p
p
ly
in
g
ML
-
GAT
to
p
r
ed
ict
tr
en
d
s
f
o
r
h
u
n
d
r
e
d
s
o
f
s
to
ck
s
in
th
e
S&
P
5
0
0
an
d
C
SI
3
0
0
in
d
ices,
th
e
r
e
s
ea
r
ch
er
s
f
o
u
n
d
it
s
ig
n
if
ican
tly
o
u
tp
er
f
o
r
m
e
d
f
iv
e
p
o
p
u
lar
b
aselin
e
ap
p
r
o
ac
h
es
.
T
h
e
m
o
d
el
ac
h
iev
e
d
s
u
b
s
tan
tial
im
p
r
o
v
em
en
ts
in
F1
-
s
co
r
e,
ac
cu
r
ac
y
,
av
er
a
g
e
d
aily
r
etu
r
n
,
an
d
s
h
ar
p
e
r
atio
,
d
em
o
n
s
tr
atin
g
th
e
ef
f
ec
tiv
en
ess
o
f
th
e
s
o
p
h
is
ticated
s
in
g
le
-
m
o
d
el
ar
c
h
itectu
r
e.
Similar
ly
,
W
an
g
et
a
l.
[
1
9
]
p
r
o
p
o
s
ed
a
n
o
v
el
f
o
r
ec
ast
m
o
d
el
n
am
e
d
s
p
ec
ial
g
ated
r
ec
u
r
r
en
t
u
n
it
(
SGR
U
)
-
atten
tio
n
m
ec
h
an
is
m
(
AM
)
,
b
ased
o
n
a
SGR
U
co
m
b
in
ed
with
an
AM
.
T
h
is
m
o
d
el
was
ap
p
lied
to
f
o
r
ec
ast
C
h
in
a's
g
o
ld
f
u
tu
r
es
p
r
ices
u
s
in
g
Sh
an
g
h
ai
Fu
tu
r
es
E
x
ch
an
g
e
d
ata
f
r
o
m
2
0
0
8
to
2
0
2
1
.
C
o
m
p
a
r
ativ
e
ex
p
er
im
en
tal
r
esu
lts
d
em
o
n
s
t
r
ated
th
at
th
e
p
r
o
p
o
s
ed
SGR
U
-
AM
m
o
d
el
ac
h
iev
ed
s
u
p
er
io
r
p
er
f
o
r
m
a
n
ce
in
b
o
th
f
o
r
ec
ast ac
cu
r
ac
y
an
d
e
f
f
icien
cy
co
m
p
a
r
ed
to
t
h
e
b
aselin
e
m
eth
o
d
s
ev
alu
ated
.
R
ec
o
g
n
izin
g
t
h
e
d
is
tin
ct
ad
v
a
n
tag
es
o
f
d
if
f
er
en
t
ar
ch
itectu
r
es,
a
s
ig
n
if
ican
t
r
ec
e
n
t
tr
en
d
f
o
cu
s
es
o
n
d
ev
elo
p
in
g
h
y
b
r
id
d
ee
p
lear
n
in
g
m
o
d
els,
p
ar
ticu
lar
l
y
co
m
b
i
n
in
g
th
e
f
ea
tu
r
e
ex
tr
ac
ti
o
n
p
o
wer
o
f
C
NNs
with
th
e
s
eq
u
en
tial
m
o
d
elin
g
s
tr
en
g
th
s
o
f
R
NN
v
ar
ian
ts
f
o
r
g
o
ld
p
r
ice
f
o
r
ec
asti
n
g
.
Fo
r
in
s
tan
c
e,
L
iv
ier
is
et
a
l.
[
8
]
ex
p
licitly
p
r
o
p
o
s
ed
a
n
ew
h
y
b
r
id
d
ee
p
lear
n
in
g
m
o
d
el
f
o
r
p
r
e
d
ictin
g
g
o
ld
p
r
ice
an
d
m
o
v
em
en
t,
aim
in
g
t
o
ca
p
italize
o
n
th
e
c
o
m
p
lem
e
n
ta
r
y
s
tr
en
g
th
s
o
f
d
if
f
er
en
t
ar
ch
it
ec
tu
r
es.
T
h
e
m
o
d
el
was
d
esig
n
ed
to
ex
p
lo
it
C
NN
f
o
r
ex
tr
ac
tin
g
r
elev
an
t
f
ea
t
u
r
es
f
r
o
m
tim
e
-
s
er
ies
d
ata
an
d
L
STM
lay
er
s
f
o
r
id
en
tify
in
g
tem
p
o
r
al
d
ep
en
d
e
n
cies.
Pre
lim
in
ar
y
e
x
p
er
im
en
tal
an
aly
s
is
s
u
g
g
ested
th
at
co
m
b
i
n
in
g
L
STM
with
c
o
n
v
o
lu
ti
o
n
al
lay
er
s
o
f
f
er
ed
a
s
ig
n
if
ican
t b
o
o
s
t to
f
o
r
ec
asti
n
g
p
er
f
o
r
m
a
n
ce
co
m
p
ar
ed
to
s
tate
-
of
-
t
h
e
-
ar
t b
aselin
e
m
o
d
els.
Hig
h
lig
h
tin
g
g
o
l
d
'
s
r
o
le
as
a
h
ed
g
e
in
v
estme
n
t,
in
an
o
th
er
ef
f
o
r
t,
B
illah
an
d
Das
[
2
0
]
p
r
o
p
o
s
ed
a
h
y
b
r
id
m
eth
o
d
c
o
m
b
in
i
n
g
1
D
-
C
NN
with
b
id
ir
ec
tio
n
al
g
ate
d
r
ec
u
r
r
en
t
u
n
its
(
B
i
-
GR
U)
f
o
r
f
o
r
ec
asti
n
g
g
o
ld
p
r
ices.
T
h
e
au
th
o
r
s
ass
er
ted
th
at
wh
ile
v
ar
io
u
s
h
y
b
r
id
a
n
d
s
tan
d
alo
n
e
m
o
d
els
p
r
o
v
id
e
d
s
atis
f
ac
to
r
y
r
esu
lts
,
th
eir
p
r
o
p
o
s
ed
1
D
-
C
NN
-
B
iG
R
U
ap
p
r
o
ac
h
p
r
o
v
e
d
m
o
r
e
r
el
iab
le.
T
h
eir
ex
p
er
im
en
ts
s
h
o
wed
it
o
u
tp
er
f
o
r
m
e
d
o
th
er
test
ed
n
etwo
r
k
s
(
in
clu
d
in
g
C
NN
-
g
ated
r
ec
u
r
r
en
t
u
n
it
(
GR
U)
,
C
NN
-
L
STM
,
C
NN
-
R
NN,
an
d
s
tan
d
alo
n
e
v
ar
ian
ts
)
b
ased
o
n
m
u
ltip
le
ev
alu
atio
n
m
etr
ics
lik
e
m
ea
n
ab
s
o
lu
te
er
r
o
r
(
MA
E
)
,
R
MSE
,
an
d
ac
h
iev
e
d
d
eter
m
in
atio
n
(
R
²)
s
co
r
e
o
f
o
v
er
9
3
%.
Fu
r
th
er
in
n
o
v
atio
n
s
in
d
ata
r
ep
r
esen
tatio
n
in
clu
d
e
in
th
e
wo
r
k
[
2
1
]
,
wh
er
e
g
o
ld
p
r
ice
d
ata
was
tr
an
s
f
o
r
m
ed
in
to
im
a
g
es
u
s
in
g
th
e
g
r
am
ian
an
g
u
lar
f
ield
(
GAF)
tech
n
iq
u
e
b
ef
o
r
e
b
ei
n
g
p
r
o
ce
s
s
ed
b
y
a
C
NN
-
L
STM
n
etwo
r
k
.
B
ey
o
n
d
s
tr
u
ctu
r
al
h
y
b
r
id
s
,
r
esear
c
h
h
as
also
in
te
g
r
ated
ex
ter
n
al
f
ac
to
r
s
to
r
ef
in
e
p
r
ed
ictio
n
s
.
Fo
r
in
s
tan
ce
,
Var
ela
et
a
l.
[
2
2
]
i
n
co
r
p
o
r
ated
Fin
B
E
R
T
-
b
ased
s
en
tim
en
t
an
aly
s
is
o
f
f
in
an
cial
n
ews
with
h
ig
h
-
f
r
eq
u
e
n
cy
tec
h
n
ical
in
d
icato
r
s
in
to
an
L
STM
f
r
am
ewo
r
k
,
wh
ich
im
p
r
o
v
ed
r
ea
l
-
tim
e
tr
ad
in
g
d
ec
is
io
n
s
f
o
r
g
o
ld
/USD e
x
ch
a
n
g
e
r
ates.
A
p
ar
allel
d
ev
elo
p
m
en
t
in
th
e
liter
atu
r
e
f
o
c
u
s
es
o
n
“
d
iv
id
e
-
an
d
-
co
n
q
u
e
r
”
s
tr
ateg
ies,
wh
er
e
co
m
p
lex
p
r
ice
s
ig
n
als
ar
e
d
ec
o
m
p
o
s
ed
in
to
s
im
p
ler
co
m
p
o
n
en
ts
b
e
f
o
r
e
m
o
d
elin
g
.
L
ian
g
et
a
l.
[
2
3
]
in
tr
o
d
u
ce
d
th
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Hyb
r
id
d
ee
p
lea
r
n
in
g
mo
d
el
f
o
r
en
h
a
n
ce
d
s
h
o
r
t
-
term g
o
ld
p
r
ice
…
(
Ho
a
n
g
Ha
N
g
u
ye
n
)
3231
I
L
C
C
m
o
d
el,
wh
ich
co
m
b
in
e
d
im
p
r
o
v
ed
c
o
m
p
lete
en
s
em
b
le
em
p
ir
ical
m
o
d
e
d
ec
o
m
p
o
s
itio
n
with
ad
ap
tiv
e
n
o
is
e
(
I
C
E
E
MD
AN)
d
ec
o
m
p
o
s
itio
n
with
a
jo
in
t
L
STM
-
C
N
N
-
co
n
v
o
l
u
tio
n
al
b
lo
ck
atten
ti
o
n
m
o
d
u
le
(
C
B
AM
)
ar
ch
itectu
r
e.
T
h
eir
r
esu
lts
in
d
icate
d
th
at
th
e
in
teg
r
atio
n
o
f
f
ea
tu
r
e
en
h
a
n
ce
m
en
t
v
ia
th
e
C
B
AM
s
ig
n
if
ican
tly
in
cr
ea
s
ed
p
r
ed
ictio
n
p
r
ec
is
io
n
.
T
h
e
ef
f
ec
tiv
en
ess
o
f
th
ese
d
ec
o
m
p
o
s
itio
n
tech
n
iq
u
es
was
f
u
r
th
er
v
alid
ated
b
y
So
n
g
et
a
l.
[
2
4
]
,
w
h
o
u
tili
ze
d
a
VM
D
-
en
s
em
b
le
em
p
ir
ical
m
o
d
e
d
ec
o
m
p
o
s
itio
n
(
EEMD
)
tech
n
iq
u
e
ac
r
o
s
s
v
ar
i
o
u
s
f
u
t
u
r
es
co
n
tr
ac
ts
(
in
clu
d
in
g
g
o
l
d
an
d
zin
c)
.
T
h
eir
em
p
ir
ical
ev
i
d
en
ce
s
h
o
we
d
th
at
m
u
ltip
le
p
r
ice
d
ec
o
m
p
o
s
itio
n
s
s
ig
n
if
ican
tly
r
e
d
u
c
e
d
e
r
r
o
r
m
e
t
r
i
cs
,
s
u
c
h
as
M
A
E
a
n
d
R
MS
E
,
c
o
m
p
a
r
e
d
t
o
s
i
n
g
l
e
-
d
e
c
o
m
p
o
s
i
t
i
o
n
m
o
d
el
s
.
F
i
n
a
ll
y
,
t
o
a
d
d
r
e
s
s
l
i
m
i
t
at
i
o
n
s
o
f
p
o
i
n
t
-
d
a
t
a
m
o
d
e
l
s
,
Y
a
n
g
e
t
a
l
.
[
2
5
]
d
e
v
e
l
o
p
e
d
i
n
t
e
r
v
a
l
-
v
a
l
u
e
d
f
o
r
e
c
a
s
t
i
n
g
s
y
s
te
m
f
o
r
n
o
n
-
f
e
r
r
o
u
s
m
e
t
a
ls
.
T
h
is
s
y
s
t
e
m
u
ti
l
iz
e
d
a
t
w
o
-
s
t
a
g
e
e
n
s
e
m
b
le
l
ea
r
n
i
n
g
a
p
p
r
o
a
c
h
w
i
t
h
t
e
m
p
o
r
a
l
f
u
s
i
o
n
t
r
a
n
s
f
o
r
m
e
r
s
a
n
d
AM
s
,
p
r
o
v
i
d
i
n
g
m
o
r
e
r
o
b
u
s
t
i
n
s
i
g
h
ts
i
n
t
o
p
r
i
c
e
r
a
n
g
es
a
n
d
m
ar
k
e
t
r
i
s
k
s
t
h
a
n
t
r
a
d
it
i
o
n
a
l
m
e
t
h
o
d
s
.
I
n
ess
en
ce
,
th
e
r
e
v
iewe
d
liter
atu
r
e
s
h
o
ws
a
clea
r
ev
o
lu
ti
o
n
f
r
o
m
tr
ad
itio
n
al
s
tatis
tica
l
m
eth
o
d
s
to
war
d
s
m
ac
h
in
e
lear
n
in
g
a
n
d
in
cr
ea
s
in
g
ly
s
o
p
h
is
ticated
d
ee
p
lear
n
in
g
m
o
d
els
f
o
r
g
o
l
d
p
r
ice
f
o
r
ec
asti
n
g
,
with
h
y
b
r
id
a
r
ch
itectu
r
es
co
m
b
in
in
g
C
NN
an
d
R
NN
elem
en
ts
s
h
o
win
g
p
ar
ticu
lar
p
r
o
m
is
e.
Desp
ite
th
is
p
r
o
g
r
ess
,
a
f
o
c
u
s
ed
co
m
p
ar
is
o
n
ev
alu
atin
g
1
D
-
C
NN
p
air
e
d
s
p
ec
if
ically
with
s
tan
d
ar
d
R
NN,
L
STM
,
an
d
B
iLST
M
f
o
r
s
h
o
r
t
-
ter
m
g
o
ld
p
r
ice
p
r
ed
ictio
n
u
n
d
er
v
ar
io
u
s
in
p
u
t w
in
d
o
w
co
n
f
ig
u
r
atio
n
s
r
em
ain
s
a
g
ap
in
th
e
cu
r
r
en
t r
esear
ch
.
T
h
is
s
tu
d
y
d
i
r
ec
tly
ad
d
r
ess
es
th
is
g
ap
b
y
p
r
o
p
o
s
in
g
a
n
d
r
ig
o
r
o
u
s
ly
ev
al
u
a
tin
g
th
e
STGP
-
Net
(
1
D
-
C
NN
-
L
STM
)
m
o
d
el
ag
ai
n
s
t th
ese
s
p
ec
if
ic
h
y
b
r
id
c
o
u
n
t
er
p
ar
ts
ac
r
o
s
s
d
if
f
er
e
n
t e
x
p
e
r
im
en
tal
s
ettin
g
s
.
3.
T
H
E
SH
O
R
T
-
T
E
RM
G
O
L
D
P
RE
DI
CT
I
O
N
NE
T
WO
R
K
M
O
DE
L
ARCH
I
T
E
CT
U
RE
T
h
is
s
ec
tio
n
d
etails
th
e
m
eth
o
d
o
lo
g
y
em
p
lo
y
ed
f
o
r
s
h
o
r
t
-
t
er
m
g
o
ld
p
r
ice
f
o
r
ec
asti
n
g
.
I
t
b
eg
in
s
b
y
in
tr
o
d
u
cin
g
th
e
o
v
er
all
ar
c
h
itectu
r
e
o
f
t
h
e
STGP
-
Net
m
o
d
el,
wh
ich
in
teg
r
ates
1
D
-
C
NN
an
d
L
STM
n
etwo
r
k
s
.
Su
b
s
eq
u
en
tly
,
th
e
s
lid
in
g
win
d
o
w
tech
n
i
q
u
e
u
s
ed
t
o
p
r
e
p
ar
e
th
e
tim
e
s
er
ies
d
ata
in
t
o
ap
p
r
o
p
r
iate
in
p
u
t
s
eq
u
en
ce
s
f
o
r
th
is
m
o
d
el
is
d
escr
ib
ed
.
Fo
llo
win
g
th
is
,
th
e
in
d
i
v
id
u
al
co
m
p
o
n
e
n
ts
o
f
th
e
h
y
b
r
id
m
o
d
e
l
(
1
D
-
C
NN
an
d
L
STM
)
ar
e
elab
o
r
ated
u
p
o
n
.
Fin
ally
,
th
e
alg
o
r
ith
m
th
at
u
tili
ze
s
th
e
p
r
o
p
o
s
ed
m
o
d
el
f
o
r
p
r
ed
ictio
n
is
p
r
esen
te
d
.
Fig
u
r
e
1
illu
s
tr
ates
th
e
p
r
o
p
o
s
ed
STGP
-
Net
ar
ch
itectu
r
e
d
esig
n
ed
f
o
r
s
h
o
r
t
-
ter
m
g
o
ld
p
r
ice
p
r
ed
ictio
n
.
I
n
p
u
t
s
eq
u
e
n
ce
s
,
cr
ea
ted
f
r
o
m
h
is
to
r
ical
g
o
ld
p
r
ice
d
ata
u
s
in
g
a
s
lid
in
g
wi
n
d
o
w
tech
n
iq
u
e,
ar
e
in
itially
f
ed
in
to
t
h
e
1
D
-
C
NN
co
m
p
o
n
en
t.
T
h
is
m
o
d
u
le
co
m
p
r
is
es
an
in
p
u
t
lay
er
f
o
llo
w
ed
b
y
th
r
ee
b
lo
ck
s
,
ea
ch
co
n
tain
in
g
a
co
n
v
o
lu
tio
n
al
lay
er
an
d
a
p
o
o
lin
g
lay
e
r
,
to
ex
tr
ac
t
h
ier
a
r
ch
ical
tem
p
o
r
al
f
ea
tu
r
es
an
d
p
atter
n
s
with
in
th
e
in
p
u
t
win
d
o
w.
T
h
e
r
esu
ltin
g
f
ea
tu
r
e
m
ap
s
f
r
o
m
t
h
e
C
NN
m
o
d
u
le
s
er
v
e
as
in
p
u
t
to
L
STM
lay
er
,
wh
ich
is
task
ed
with
m
o
d
elin
g
lo
n
g
er
-
r
a
n
g
e
s
eq
u
en
tial d
ep
en
d
en
cies p
r
esen
t in
th
e
e
x
tr
ac
ted
f
ea
tu
r
es.
Fig
u
r
e
1
.
STGP
-
Net
ar
ch
itectu
r
e
Fin
ally
,
th
e
o
u
tp
u
t
s
eq
u
en
ce
f
r
o
m
th
e
L
STM
lay
er
is
p
r
o
ce
s
s
ed
b
y
a
f
u
lly
co
n
n
ec
ted
(
d
e
n
s
e)
o
u
tp
u
t
lay
er
.
T
h
is
ty
p
e
o
f
lay
er
,
wh
e
r
e
ea
ch
n
eu
r
o
n
r
ec
eiv
es
in
p
u
t
f
r
o
m
all
n
e
u
r
o
n
s
in
th
e
p
r
ev
io
u
s
lay
er
,
ef
f
ec
tiv
ely
m
ap
s
th
e
lear
n
ed
h
ig
h
-
lev
el
s
eq
u
en
tial
f
ea
tu
r
es
to
t
h
e
f
in
al
p
r
ed
icted
g
o
ld
p
r
ice(
s
)
.
T
h
is
a
r
ch
itectu
r
e
aim
s
to
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t
20
26
:
3
2
2
8
-
3
2
3
9
3232
en
h
an
ce
p
r
e
d
ictio
n
p
er
f
o
r
m
a
n
ce
b
y
co
m
b
in
in
g
th
e
1
D
-
C
NN'
s
p
r
o
f
icien
cy
in
lo
ca
l
p
atter
n
ex
tr
ac
tio
n
with
th
e
L
STM
's ca
p
ab
ilit
y
f
o
r
m
o
d
eli
n
g
tem
p
o
r
al
d
y
n
am
ics.
3
.
1
.
T
im
e
s
er
ies s
eg
m
ent
a
t
i
o
n us
ing
s
lid
i
ng
wind
o
ws
Giv
en
th
e
h
is
to
r
ical
tr
an
s
ac
tio
n
al
d
ataset
o
f
g
o
ld
as
=
{
|
∈
;
=
0
,
±
1
,
±
2
,
⋯
}
.
T
h
is
tim
e
s
er
ies
d
ataset
(
∈
×
)
m
ain
tain
s
th
e
ch
r
o
n
o
lo
g
ical
o
r
d
er
o
f
o
b
s
er
v
atio
n
s
.
E
ac
h
d
ata
p
o
in
t
is
d
e
n
o
ted
as
∈
r
ep
r
esen
ts
th
e
p
r
ices
o
f
a
g
o
ld
tr
an
s
ac
tio
n
at
a
s
p
ec
if
ic
ti
m
e
t
.
T
o
p
r
ep
a
r
e
th
is
s
eq
u
en
t
ial
d
ata
f
o
r
in
p
u
t
in
t
o
o
u
r
p
r
e
d
ictiv
e
m
o
d
el,
a
s
lid
in
g
win
d
o
w
m
eth
o
d
is
ap
p
lied
.
T
h
is
tech
n
i
q
u
e
ex
tr
ac
ts
f
ix
ed
-
s
ize
s
u
b
s
eq
u
en
ce
s
(
win
d
o
ws)
b
y
m
o
v
in
g
ac
r
o
s
s
th
e
o
r
ig
in
al
d
a
taset.
T
h
e
m
ath
em
atica
l
d
ef
in
itio
n
o
f
th
e
s
lid
in
g
win
d
o
w
m
eth
o
d
is
p
r
esen
ted
i
n
(
1
)
.
(
,
)
:
×
→
×
×
,
∀
∈
(
1
)
W
h
er
e
∈
×
,
w
is
th
e
win
d
o
w
s
ize
,
an
d
≪
.
I
n
s
h
o
r
t,
th
e
f
u
n
ctio
n
(
,
)
m
ap
s
th
e
o
r
ig
in
al
d
ataset
f
r
o
m
a
te
n
s
o
r
with
r
an
k
2
(
×
)
to
a
ten
s
o
r
with
r
an
k
3
(
×
×
)
.
T
h
e
ten
s
o
r
r
esu
ltin
g
f
r
o
m
th
e
s
lid
in
g
win
d
o
w
o
p
er
atio
n
is
well
-
s
u
ited
f
o
r
s
u
b
s
eq
u
en
t
p
r
o
ce
s
s
in
g
.
I
t
ai
d
s
m
o
d
el
tr
ain
in
g
a
n
d
f
ea
tu
r
e
en
g
in
ee
r
in
g
,
ass
is
ts
in
ca
p
tu
r
in
g
s
h
o
r
t
-
te
r
m
tem
p
o
r
al
p
atter
n
s
,
an
d
ca
n
co
n
tr
ib
u
te
to
r
e
d
u
ce
d
co
m
p
u
ta
tio
n
.
T
h
e
s
p
ec
if
ic
im
p
lem
en
t
atio
n
o
f
th
e
s
lid
in
g
win
d
o
w
p
r
o
ce
s
s
o
n
th
e
tim
e
s
er
ies d
ata
is
o
u
tlin
ed
in
Alg
o
r
i
th
m
1
.
Alg
o
r
ith
m
1
.
Sli
d
in
g
win
d
o
w
function slideWindow(data, window_size, forecast_size)
data_size
←
|
dat
a
|
X
←
[
]
y
←
[
]
for i
from window_size to data_size
–
window_size
–
forecast_size:
X
←
X
∪
dat
a
[
i
−
wi
n
do
w
_
si
z
e
−
f
o
r
e
c
a
st
_
si
z
e
∶
i
−
f
o
r
e
c
a
st
_
si
z
e
,
∶
]
y
←
y
∪
dat
a
[
i
−
f
o
r
e
c
a
st
_
si
z
e
∶
i
,
∶
]
return X, y
end function
3
.
2
.
O
ne
-
dim
ens
io
na
l c
o
nv
o
lutio
na
l neura
l net
wo
rk
co
mp
o
nent
f
o
r
f
ea
t
ure
ex
t
ra
ct
io
n
1D
-
C
NN
is
a
s
p
ec
ialized
C
N
N
ar
ch
itectu
r
e
tailo
r
e
d
f
o
r
s
e
q
u
en
tial
d
ata,
lik
e
tim
e
s
er
ies
o
r
n
at
u
r
al
lan
g
u
ag
e,
f
o
cu
s
in
g
o
n
id
en
ti
f
y
in
g
p
atter
n
s
alo
n
g
th
e
tem
p
o
r
al
d
im
en
s
io
n
.
A
ty
p
ical
1
D
-
C
NN
co
m
p
r
is
es
s
ev
er
al
k
ey
lay
e
r
s
:
co
n
v
o
l
u
tio
n
al
lay
er
s
ar
e
p
r
im
ar
ily
r
esp
o
n
s
ib
le
f
o
r
e
x
tr
ac
tin
g
r
ele
v
an
t
lo
ca
l
f
ea
tu
r
es
f
r
o
m
th
e
in
p
u
t
s
eq
u
en
ce
s
.
Activ
ati
o
n
f
u
n
ctio
n
s
ap
p
lied
with
in
t
h
ese
lay
er
s
in
tr
o
d
u
ce
n
o
n
-
lin
ea
r
ity
,
en
ab
lin
g
t
h
e
n
etwo
r
k
to
m
o
d
el
co
m
p
lex
p
a
tter
n
s
.
Po
o
lin
g
la
y
er
s
ar
e
c
o
m
m
o
n
ly
i
n
clu
d
ed
to
r
ed
u
ce
th
e
d
im
en
s
io
n
ality
o
f
th
e
f
ea
tu
r
e
m
a
p
s
,
wh
ich
ca
n
i
m
p
r
o
v
e
co
m
p
u
tatio
n
al
ef
f
icien
cy
an
d
r
o
b
u
s
tn
ess
.
I
n
th
e
c
o
n
tex
t
o
f
th
is
r
esear
ch
,
th
e
1
D
-
C
NN
m
o
d
u
le
is
in
teg
r
ated
with
an
L
STM
m
o
d
u
le.
Fo
r
m
ally
,
a
1
D
c
o
n
v
o
lu
tio
n
al
o
p
er
atio
n
ca
lc
u
lates
an
o
u
tp
u
t
f
ea
tu
r
e
(
)
at
tim
estep
t
b
y
ap
p
ly
in
g
a
k
er
n
el
w
o
f
s
ize
k
to
a
co
r
r
esp
o
n
d
in
g
s
eg
m
en
t o
f
th
e
in
p
u
t s
eq
u
en
ce
x
,
as sh
o
wn
in
(
2
)
.
(
)
=
(
∗
)
(
)
=
∑
(
+
)
∙
(
)
−
1
=
0
(
2
)
W
h
er
e
∗
d
en
o
tes
th
e
co
n
v
o
lu
ti
o
n
o
p
er
atio
n
,
(
+
)
is
th
e
(
+
)
ℎ
in
p
u
t
elem
en
t,
an
d
(
)
is
th
e
−
ℎ
elem
en
t
o
f
t
h
e
k
er
n
el.
T
h
e
r
e
ctif
ied
lin
ea
r
u
n
it
(
R
eL
U)
s
er
v
es
as
th
e
ac
tiv
atio
n
f
u
n
ctio
n
with
in
th
is
h
y
b
r
id
m
o
d
el;
its
f
o
r
m
u
la
is
p
r
esen
ted
in
(
3
)
.
(
)
=
ma
x
(
0
,
)
(
3
)
T
h
e
m
ax
p
o
o
lin
g
o
p
e
r
atio
n
,
u
t
ilized
with
in
th
e
h
y
b
r
id
m
o
d
el
,
is
f
o
r
m
ally
d
e
f
in
ed
b
y
(
4
)
.
(
)
=
ma
x
(
(
)
,
(
+
1
)
,
⋯
,
(
+
−
1
)
)
(
4
)
T
h
e
1
D
-
C
NN
m
o
d
u
le
is
tr
ain
ed
b
y
a
d
ju
s
tin
g
its
f
ilter
weig
h
ts
v
ia
b
ac
k
p
r
o
p
ag
atio
n
.
A
lo
s
s
f
u
n
ctio
n
f
ir
s
t
q
u
an
tifie
s
th
e
er
r
o
r
b
etwe
en
th
e
p
r
e
d
icted
an
d
ac
tu
al
o
u
tp
u
ts
.
B
ac
k
p
r
o
p
ag
atio
n
th
en
ca
lcu
lates
th
e
g
r
ad
ien
ts
o
f
th
is
lo
s
s
r
elativ
e
to
th
e
n
etwo
r
k
'
s
weig
h
ts
.
T
h
ese
g
r
ad
ien
ts
g
u
id
e
th
e
w
eig
h
t
u
p
d
ate
s
tep
,
ty
p
ically
ex
ec
u
ted
u
s
in
g
th
e
ad
ap
tiv
e
m
o
m
e
n
t
esti
m
atio
n
(
Ad
am
)
o
p
tim
izer
,
to
m
i
n
im
i
ze
th
e
er
r
o
r
.
T
h
e
s
p
ec
if
ic
ca
lcu
latio
n
f
o
r
th
e
co
n
v
o
lu
tio
n
al
lay
er
g
r
ad
ien
ts
is
d
ef
in
ed
in
(
5
)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Hyb
r
id
d
ee
p
lea
r
n
in
g
mo
d
el
f
o
r
en
h
a
n
ce
d
s
h
o
r
t
-
term g
o
ld
p
r
ice
…
(
Ho
a
n
g
Ha
N
g
u
ye
n
)
3233
(
)
=
∑
(
)
∙
(
+
)
(
5
)
W
h
er
e
(
)
=
∑
(
+
)
∙
(
)
−
1
=
0
an
d
(
)
is
th
e
g
r
ad
ien
t
o
f
th
e
lo
s
s
with
r
esp
ec
t to
th
e
o
u
t
p
u
t.
3
.
3
.
L
o
ng
s
ho
rt
-
t
er
m
m
emo
ry
co
m
po
nent
f
o
r
t
e
m
po
ra
l
m
o
delin
g
I
n
th
is
h
y
b
r
id
m
o
d
el,
th
e
L
STM
lay
er
,
co
m
p
o
s
ed
o
f
m
u
lti
p
le
L
STM
ce
lls
,
is
d
esig
n
ed
to
ca
p
tu
r
e
lo
n
g
-
r
a
n
g
e
d
e
p
en
d
e
n
cies
with
in
th
e
g
o
ld
p
r
ice
s
eq
u
e
n
ce
s
an
d
to
h
elp
m
itig
ate
th
e
v
an
is
h
in
g
g
r
ad
ie
n
t
p
r
o
b
lem
co
m
m
o
n
in
s
im
p
ler
r
ec
u
r
r
en
t
n
etwo
r
k
s
.
T
h
e
s
p
ec
if
ic
in
ter
n
al
s
tr
u
ctu
r
e
o
f
an
L
STM
ce
ll
is
v
is
u
alize
d
in
Fig
u
r
e
2
.
T
h
e
co
r
e
o
f
an
L
ST
M
ce
ll
in
clu
d
es
f
o
r
g
et,
in
p
u
t,
an
d
o
u
tp
u
t
g
ates
th
at
m
an
ag
e
its
s
ta
te.
T
h
e
f
o
r
g
et
g
ate
s
elec
tiv
ely
r
em
o
v
es
in
f
o
r
m
atio
n
f
r
o
m
t
h
e
p
r
ev
i
o
u
s
ce
ll
s
tate
(
)
,
wh
ile
th
e
in
p
u
t
g
ate
ad
d
s
r
elev
an
t
n
ew
in
f
o
r
m
atio
n
.
T
h
e
o
u
tp
u
t
g
ate
d
eter
m
in
es
th
e
n
ex
t
h
id
d
en
s
t
ate
(
ℎ
)
-
th
e
ce
ll
's
o
u
tp
u
t
f
o
r
th
e
tim
estep
-
b
y
f
ilter
in
g
th
e
u
p
d
ated
.
T
h
ese
o
p
er
atio
n
s
ca
n
b
e
f
o
r
m
ally
d
e
s
cr
ib
ed
b
y
th
e
tu
p
le
ℒ
=
〈
,
,
,
,
,
̃
ℎ
〉
,
wh
o
s
e
co
m
p
o
n
en
ts
ar
e
d
etaile
d
in
T
ab
le
1
.
Fig
u
r
e
2
.
An
L
STM
ce
ll st
r
u
ctu
r
e
T
ab
le
1
.
Ma
th
em
atica
l d
escr
ip
tio
n
o
f
L
STM
ce
ll c
o
m
p
o
n
en
t
s
M
a
t
h
sy
m
b
o
l
s
M
e
a
n
i
n
g
(
)
Th
e
i
n
p
u
t
v
e
c
t
o
r
t
o
t
h
e
LS
TM
c
e
l
l
Th
e
i
n
p
u
t
-
w
e
i
g
h
t
m
a
t
r
i
x
w
h
e
r
e
d
c
a
n
b
e
f
,
i
,
o
,
o
r
c.
Th
e
ma
t
r
i
x
o
f
t
h
e
r
e
c
u
r
r
e
n
t
c
o
n
n
e
c
t
i
o
n
s w
h
e
r
e
d
c
a
n
b
e
f
,
i
,
o
o
r
c.
Th
e
b
i
a
s
v
e
c
t
o
r
w
h
e
r
e
d
c
a
n
b
e
f
,
i
,
o
,
o
r
c.
Th
e
si
g
m
o
i
d
f
u
n
c
t
i
o
n
Th
e
h
y
p
e
r
b
o
l
i
c
t
a
n
g
e
n
t
f
u
n
c
t
i
o
n
=
(
(
)
+
ℎ
(
−
1
)
+
)
Th
e
f
o
r
g
e
t
g
a
t
e
o
f
t
h
e
mem
o
r
y
c
e
l
l
c
a
t
t
i
me
t
.
=
(
(
)
+
ℎ
(
−
1
)
+
)
Th
e
i
n
p
u
t
g
a
t
e
o
f
t
h
e
me
mo
r
y
c
e
l
l
c
a
t
t
i
m
e
t
.
=
(
(
)
+
ℎ
(
−
1
)
+
)
Th
e
o
u
t
p
u
t
g
a
t
e
o
f
t
h
e
mem
o
r
y
c
e
l
l
c
at
t
i
m
e
t
.
(
)
̃
=
(
(
)
+
ℎ
(
−
1
)
+
)
Th
e
c
e
l
l
i
n
p
u
t
a
c
t
i
v
a
t
i
o
n
v
e
c
t
o
r
a
t
t
h
e
t
i
m
e
st
e
p
t
.
(
)
=
⨀
(
−
1
)
+
⨀
(
)
̃
Th
e
v
e
c
t
o
r
a
t
t
h
e
t
i
me
st
e
p
t
.
ℎ
(
)
=
⨀
(
(
)
)
Th
e
o
u
t
p
u
t
v
e
c
t
o
r
o
f
t
h
e
LS
TM
c
e
l
l
(
o
r
t
h
e
ℎ
v
e
c
t
o
r
)
a
t
t
h
e
t
i
me
st
e
p
t
.
⨀
Th
i
s
o
p
e
r
a
t
o
r
d
e
n
o
t
e
s
t
h
e
e
l
e
me
n
t
-
w
i
s
e
p
r
o
d
u
c
t
.
3
.
4
.
Sh
o
rt
-
t
er
m
g
o
ld predict
io
n net
wo
rk
predict
io
n pro
ce
du
re
Alg
o
r
ith
m
2
p
r
esen
ts
th
e
m
eth
o
d
o
lo
g
y
f
o
r
s
h
o
r
t
-
ter
m
g
o
l
d
p
r
ice
f
o
r
ec
asti
n
g
u
s
in
g
th
e
d
ev
elo
p
ed
h
y
b
r
id
1
D
-
C
NN
-
L
STM
ar
ch
it
ec
tu
r
e.
T
h
e
p
r
o
ce
d
u
r
e
in
v
o
lv
e
s
p
r
o
ce
s
s
in
g
h
is
to
r
ical
g
o
ld
p
r
ice
d
ata,
ap
p
ly
in
g
a
s
lid
in
g
win
d
o
w
tec
h
n
iq
u
e
to
g
en
er
ate
in
p
u
t
s
eg
m
e
n
ts
,
an
d
lev
er
ag
in
g
th
e
tr
ain
ed
m
o
d
e
l
to
p
r
e
d
ict
p
r
ices
ac
r
o
s
s
a
d
esig
n
ated
f
o
r
ec
ast
h
o
r
izo
n
.
T
h
e
alg
o
r
ith
m
en
ca
p
s
u
lates
th
e
n
ec
ess
ar
y
d
ata
p
r
ep
r
o
ce
s
s
in
g
,
m
o
d
el
in
itializatio
n
,
an
d
p
r
ed
ictio
n
g
en
er
atio
n
s
tep
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t
20
26
:
3
2
2
8
-
3
2
3
9
3234
Alg
o
r
ith
m
2
.
STGP
-
Net
p
r
ed
i
cts n
ex
t n
-
d
a
y
g
o
l
d
p
r
ices
function predict_Next_N_Prices(file, window_size, forecast_size)
timeseries_data
←
read_data(file)
tensors_data
←
slideWindow(timeseries_data, window_size, forecast_size)
model
←
initialize_hybrid_model()
results
←
model.predict(tensors_data.X)
return results
end function
Sp
ec
if
ically
,
Alg
o
r
ith
m
2
b
e
g
in
s
b
y
lo
ad
i
n
g
th
e
h
is
to
r
ical
g
o
ld
p
r
ice
d
ata.
T
h
is
d
at
a
is
th
en
p
r
o
ce
s
s
ed
b
y
th
e
s
lid
eWi
n
d
o
w
f
u
n
ctio
n
,
u
s
in
g
s
p
ec
if
ied
wi
n
d
o
w_
s
ize
an
d
f
o
r
ec
ast_
s
ize
p
ar
am
eter
s
,
to
cr
ea
te
ap
p
r
o
p
r
iately
s
h
ap
ed
in
p
u
t
ten
s
o
r
s
.
Nex
t,
th
e
p
r
e
-
tr
ain
e
d
STGP
-
Net
m
o
d
el
is
in
itiali
ze
d
(
i.e
.
,
l
o
ad
ed
)
.
T
h
ese
p
r
ep
ar
ed
in
p
u
t
ten
s
o
r
s
ar
e
f
ed
in
to
th
e
m
o
d
el'
s
p
r
ed
ict
m
eth
o
d
to
g
e
n
er
ate
th
e
p
r
ed
icted
g
o
ld
p
r
ice
v
alu
es
f
o
r
th
e
n
e
x
t
f
o
r
ec
ast_
s
ize
tim
e
s
tep
s
.
Fin
ally
,
th
e
al
g
o
r
ith
m
r
et
u
r
n
s
th
ese
p
r
ed
icte
d
r
esu
lts
,
p
r
o
v
id
i
n
g
th
e
s
h
o
r
t
-
ter
m
f
o
r
ec
ast
f
o
r
g
o
ld
p
r
ices.
T
h
is
o
u
tlin
es
an
ef
f
icien
t
s
eq
u
en
ce
o
f
d
ata
p
r
ep
ar
atio
n
,
m
o
d
el
ex
ec
u
tio
n
,
a
n
d
f
o
r
ec
ast g
en
e
r
a
tio
n
.
4.
E
XP
E
R
I
M
E
N
T
T
h
e
em
p
ir
ical
b
asis
f
o
r
t
h
is
r
esear
ch
is
a
co
m
p
r
eh
e
n
s
iv
e
d
ataset
o
f
h
is
to
r
ical
g
o
ld
p
r
ice
s
o
b
tain
ed
f
r
o
m
th
e
Ya
h
o
o
Fin
an
ce
s
er
v
ice
at
h
ttp
s
://fin
an
ce
.
y
ah
o
o
.
co
m
/.
Utilizin
g
th
e
y
f
in
a
n
ce
Py
th
o
n
lib
r
a
r
y
at
h
ttp
s
://p
y
p
i.o
r
g
/
p
r
o
ject/y
f
in
a
n
ce
/,
d
aily
r
e
co
r
d
s
i
n
clu
d
in
g
o
p
en
,
h
ig
h
,
lo
w,
clo
s
e,
an
d
v
o
l
u
m
e
wer
e
c
o
llected
f
o
r
t
h
e
p
e
r
io
d
s
p
an
n
in
g
A
u
g
u
s
t
3
0
,
2
0
0
0
,
t
o
Ap
r
il
8
,
2
0
2
5
.
T
h
is
d
ataset
s
er
v
ed
as
th
e
f
o
u
n
d
atio
n
f
o
r
tr
ain
in
g
an
d
ev
alu
atin
g
th
e
p
r
o
p
o
s
ed
s
h
o
r
t
-
ter
m
p
r
ed
ictio
n
m
o
d
el
s
.
All
ex
p
er
im
en
ts
wer
e
p
er
f
o
r
m
ed
with
in
th
e
Go
o
g
le
C
o
lab
en
v
ir
o
n
m
en
t
at
h
ttp
s
://co
lab
.
r
esear
ch
.
g
o
o
g
le.
co
m
/,
wh
ich
was
s
elec
ted
f
o
r
its
s
ig
n
if
ican
t
ad
v
an
tag
es,
in
clu
d
in
g
f
r
ee
ac
ce
s
s
to
p
o
wer
f
u
l
co
m
p
u
tin
g
r
eso
u
r
ce
s
lik
e
g
r
ap
h
ics
p
r
o
ce
s
s
in
g
u
n
its
(
GPU
s
)
,
ess
en
tial
f
o
r
ac
ce
ler
atin
g
d
ee
p
lear
n
in
g
m
o
d
el
tr
ain
in
g
,
an
d
its
p
r
e
-
co
n
f
ig
u
r
ed
s
etu
p
with
m
ajo
r
d
ata
s
cien
ce
an
d
m
ac
h
i
n
e
lear
n
in
g
lib
r
ar
ies
r
ea
d
ily
av
ailab
le.
T
o
p
r
ep
ar
e
th
e
r
aw
tim
e
s
er
ies
d
ata
f
o
r
in
p
u
t
in
to
th
e
h
y
b
r
id
d
ee
p
lear
n
in
g
m
o
d
els,
a
s
lid
in
g
win
d
o
w
tech
n
iq
u
e
was
em
p
lo
y
ed
.
T
h
i
s
ap
p
r
o
ac
h
t
r
an
s
f
o
r
m
s
th
e
s
eq
u
en
tial
d
ata
in
to
o
v
er
la
p
p
in
g
win
d
o
ws
o
f
f
ix
ed
s
ize,
wh
er
e
ea
ch
win
d
o
w
s
er
v
es
as
an
in
p
u
t
s
eq
u
en
ce
to
p
r
e
d
ict
a
s
u
b
s
eq
u
en
t
s
eq
u
en
ce
o
f
a
s
p
ec
if
ied
f
o
r
ec
ast
s
ize.
T
o
in
v
esti
g
ate
th
e
im
p
ac
t
o
f
v
ar
y
in
g
in
p
u
t
len
g
th
s
an
d
p
r
ed
ictio
n
h
o
r
izo
n
s
,
th
r
ee
d
is
tin
ct
co
n
f
ig
u
r
atio
n
s
wer
e
estab
lis
h
ed
.
C
o
n
f
ig
u
r
atio
n
1
u
tili
ze
d
a
win
d
o
w
s
ize
o
f
1
0
d
ay
s
to
p
r
ed
ict
th
e
n
ex
t
3
d
ay
s
.
C
o
n
f
ig
u
r
atio
n
2
u
s
ed
a
win
d
o
w
s
ize
o
f
1
5
d
ay
s
to
f
o
r
ec
as
t
th
e
s
u
b
s
eq
u
en
t
5
d
ay
s
.
Fin
ally
,
c
o
n
f
ig
u
r
atio
n
3
in
v
o
lv
ed
a
win
d
o
w
s
ize
o
f
2
0
d
ay
s
f
o
r
p
r
e
d
ictin
g
t
h
e
n
e
x
t
7
d
a
y
s
.
T
h
is
p
r
o
ce
s
s
r
esu
lte
d
in
t
h
r
ee
s
ep
ar
ate
win
d
o
w
-
b
ased
d
atasets
,
ea
ch
tailo
r
ed
to
o
n
e
o
f
th
ese
co
n
f
ig
u
r
atio
n
s
.
Fo
r
t
h
e
p
u
r
p
o
s
e
o
f
m
o
d
el
tr
ain
in
g
,
v
alid
atio
n
,
an
d
f
in
al
ev
alu
atio
n
,
ea
ch
o
f
th
ese
th
r
ee
d
atasets
was
co
n
s
i
s
ten
tly
p
ar
titi
o
n
ed
in
to
a
tr
ain
in
g
s
et
(
7
0
%),
a
v
alid
atio
n
s
et
(
1
5
%)
,
an
d
a
test
s
et
(
1
5
%).
T
h
is
7
0
%
-
15%
-
1
5
%
s
p
lit
was
a
p
p
li
ed
u
n
i
f
o
r
m
l
y
ac
r
o
s
s
all
th
r
ee
co
n
f
ig
u
r
atio
n
s
to
e
n
s
u
r
e
co
m
p
ar
ab
le
ev
al
u
atio
n
co
n
d
itio
n
s
.
Fig
u
r
e
s
3
to
5
illu
s
tr
ate
th
e
tr
ain
in
g
a
n
d
v
alid
atio
n
lo
s
s
cu
r
v
es
f
o
r
t
h
e
th
r
ee
ev
alu
ated
h
y
b
r
i
d
d
ee
p
l
ea
r
n
in
g
m
o
d
els
(
1
D
-
C
NN+
R
NN,
STGP
-
Net,
an
d
1D
-
C
NN+
B
iL
STM
)
ac
r
o
s
s
c
o
n
f
ig
u
r
atio
n
s
1
,
2
,
an
d
3
,
r
esp
e
ctiv
ely
.
Fig
u
r
e
3
.
L
o
s
s
cu
r
v
es o
f
c
o
n
f
i
g
u
r
atio
n
1
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Hyb
r
id
d
ee
p
lea
r
n
in
g
mo
d
el
f
o
r
en
h
a
n
ce
d
s
h
o
r
t
-
term g
o
ld
p
r
ice
…
(
Ho
a
n
g
Ha
N
g
u
ye
n
)
3235
Fig
u
r
e
4
.
L
o
s
s
cu
r
v
es o
f
c
o
n
f
i
g
u
r
atio
n
2
Fig
u
r
e
5
.
L
o
s
s
cu
r
v
es o
f
c
o
n
f
i
g
u
r
atio
n
3
Ob
s
er
v
in
g
th
e
lo
s
s
cu
r
v
es p
r
esen
ted
in
Fig
u
r
es
3
t
o
5
,
all
th
r
ee
h
y
b
r
id
m
o
d
els
ex
h
i
b
it
a
ch
ar
ac
ter
is
tic
r
ap
id
d
ec
r
ea
s
e
in
b
o
th
tr
ain
i
n
g
an
d
v
alid
atio
n
l
o
s
s
d
u
r
in
g
th
e
i
n
itial
ep
o
c
h
s
ac
r
o
s
s
all
co
n
f
ig
u
r
atio
n
s
,
in
d
icatin
g
s
u
cc
ess
f
u
l
lear
n
in
g
in
itiatio
n
.
Su
b
s
eq
u
en
tly
,
th
e
l
o
s
s
v
alu
es
f
o
r
b
o
th
tr
ain
in
g
an
d
v
alid
atio
n
ten
d
to
s
tab
ilize,
s
u
g
g
esti
n
g
co
n
v
er
g
e
n
ce
to
war
d
s
a
m
in
im
u
m
.
No
t
ab
ly
,
th
e
g
a
p
b
etwe
en
t
h
e
tr
ai
n
in
g
an
d
v
alid
atio
n
lo
s
s
r
em
ain
s
r
elativ
ely
s
m
all
an
d
d
o
es
n
o
t
s
ig
n
if
ican
tly
wi
d
en
af
ter
th
e
in
itial
s
tab
ilizati
o
n
p
h
ase
f
o
r
m
o
s
t
m
o
d
els
an
d
co
n
f
ig
u
r
atio
n
s
,
im
p
ly
in
g
th
at
s
u
b
s
tan
tial
o
v
e
r
f
itti
n
g
d
id
n
o
t
o
cc
u
r
with
in
t
h
e
1
0
0
tr
ain
in
g
e
p
o
ch
s
s
h
o
wn
.
T
h
is
g
en
er
al
b
eh
a
v
io
r
h
o
ld
s
ac
r
o
s
s
th
e
d
if
f
er
en
t w
in
d
o
w
an
d
f
o
r
ec
ast s
ettin
g
s
r
ep
r
esen
ted
b
y
th
e
th
r
ee
co
n
f
ig
u
r
atio
n
s
,
d
e
m
o
n
s
tr
atin
g
co
n
s
is
ten
t
lear
n
in
g
d
y
n
a
m
ics
f
o
r
th
e
e
v
alu
ated
m
o
d
els
u
n
d
e
r
th
ese
ex
p
er
im
en
tal
co
n
d
itio
n
s
.
T
o
q
u
an
titativ
ely
ass
ess
th
e
p
r
ed
ictiv
e
ac
cu
r
ac
y
an
d
g
o
o
d
n
ess
-
of
-
f
it
o
f
th
e
d
if
f
er
e
n
t
h
y
b
r
id
m
o
d
els,
f
o
u
r
s
tan
d
a
r
d
ev
alu
atio
n
m
etr
ics
wer
e
em
p
lo
y
ed
:
MA
E
in
(6
)
,
m
ea
n
s
q
u
ar
ed
e
r
r
o
r
(
MSE
)
in
(
7
)
,
R
MSE
in
(
8
)
,
an
d
th
e
R²
in
(
9
)
.
(
,
̂
)
=
1
∑
(
−
̂
)
2
−
1
=
0
(
6
)
(
,
̂
)
=
1
∑
|
−
̂
|
−
1
=
0
(
7
)
(
,
̂
)
=
√
(
,
̂
)
(
8
)
2
(
,
̂
)
=
1
−
∑
(
−
̂
)
2
−
1
=
0
∑
(
−
̅
)
2
−
1
=
0
(
9
)
W
h
er
e
N
is
th
e
to
tal
s
am
p
les
;
̂
ar
e
tr
u
e
an
d
p
r
e
d
icted
v
al
u
es,
r
esp
ec
tiv
ely
;
̅
=
1
∑
−
1
=
0
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t
20
26
:
3
2
2
8
-
3
2
3
9
3236
L
o
wer
v
alu
es
f
o
r
MA
E
,
MSE
,
an
d
R
MSE
in
d
icate
s
m
all
er
p
r
ed
ictio
n
er
r
o
r
s
,
wh
ile
an
R²
v
alu
e
clo
s
er
to
1
s
ig
n
if
ies
a
b
etter
f
it
o
f
th
e
m
o
d
el
to
th
e
ac
tu
al
d
ata.
T
h
e
co
m
p
u
ted
v
alu
e
s
f
o
r
th
ese
m
etr
ics,
co
m
p
ar
in
g
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
1
D
-
C
NN+
R
NN,
STGP
-
Net
(
1
D
-
C
NN+
L
STM
)
,
an
d
1
D
-
C
NN+
B
iL
STM
m
o
d
els ac
r
o
s
s
all
th
r
ee
ex
p
er
im
en
tal
co
n
f
ig
u
r
atio
n
s
(
c
o
n
f
ig
u
r
atio
n
s
1
,
2
,
an
d
3
)
o
n
th
e
test
s
et
.
T
h
e
r
esu
lts
ar
e
s
u
m
m
ar
ized
in
T
a
b
le
2
.
T
ab
le
2
.
Mo
d
el
p
er
f
o
r
m
a
n
ce
s
ac
r
o
s
s
th
r
ee
co
n
f
i
g
u
r
atio
n
s
C
o
n
f
i
g
u
r
a
t
i
o
n
M
o
d
e
l
M
A
E
M
S
E
R
M
S
E
R²
C
o
n
f
i
g
u
r
a
t
i
o
n
1
1D
-
C
N
N
a
n
d
R
N
N
0
.
0
0
1
8
9
9
0
.
0
0
5
7
9
9
0
.
0
4
3
5
8
3
-
0
.
0
0
2
2
9
7
S
TG
P
-
N
e
t
0
.
0
0
1
8
1
5
0
.
0
0
7
2
5
4
0
.
0
4
2
6
0
2
0
.
0
4
2
4
2
7
1D
-
C
N
N
a
n
d
B
i
LST
M
0
.
0
0
1
7
6
3
0
.
0
0
5
9
5
5
0
.
0
4
1
9
8
4
0
.
0
7
8
7
9
C
o
n
f
i
g
u
r
a
t
i
o
n
2
1D
-
C
N
N
a
n
d
R
N
N
0
.
0
0
1
1
7
0
.
0
0
4
2
7
2
0
.
0
3
4
2
0
7
0
.
1
8
8
8
3
S
TG
P
-
N
e
t
0
.
0
0
1
1
3
7
0
.
0
0
4
4
4
5
0
.
0
3
3
7
1
5
0
.
1
9
8
8
8
9
1D
-
C
N
N
a
n
d
B
i
LST
M
0
.
0
0
1
2
5
3
0
.
0
1
0
1
6
0
.
0
3
5
3
9
3
-
0
.
3
2
9
7
8
8
C
o
n
f
i
g
u
r
a
t
i
o
n
3
1D
-
C
N
N
a
n
d
R
N
N
0
.
0
0
0
9
1
4
0
.
0
0
9
3
8
8
0
.
0
3
0
2
3
2
-
0
.
7
0
5
0
8
8
S
TG
P
-
N
e
t
0
.
0
0
0
8
7
2
0
.
0
0
3
2
6
5
0
.
0
2
9
5
3
2
0
.
3
4
1
9
1D
-
C
N
N
a
n
d
B
i
LST
M
0
.
0
0
0
8
7
0
.
0
0
4
0
3
1
0
.
0
2
9
4
9
6
0
.
2
3
6
3
3
5
An
an
aly
s
is
o
f
th
e
q
u
an
titativ
e
r
esu
lts
p
r
esen
ted
in
T
ab
le
2
r
ev
ea
ls
d
is
tin
ct
p
er
f
o
r
m
an
ce
d
if
f
er
en
ce
s
am
o
n
g
th
e
ev
al
u
ated
h
y
b
r
id
m
o
d
els
ac
r
o
s
s
th
e
th
r
ee
co
n
f
ig
u
r
atio
n
s
.
W
h
ile
th
e
1
D
-
C
NN+
B
iLST
M
m
o
d
el
s
h
o
wed
th
e
lo
west
MA
E
a
n
d
R
MSE
in
c
o
n
f
ig
u
r
atio
n
1
,
its
p
er
f
o
r
m
an
ce
s
ig
n
if
i
ca
n
tly
d
eg
r
ad
e
d
in
c
o
n
f
ig
u
r
atio
n
2
,
ex
h
ib
itin
g
th
e
h
ig
h
est
e
r
r
o
r
s
a
n
d
a
n
e
g
ativ
e
R²
v
alu
e
.
C
o
n
v
er
s
ely
,
th
e
1
D
-
C
NN+
R
NN
m
o
d
el
p
er
f
o
r
m
ed
p
o
o
r
ly
i
n
c
o
n
f
ig
u
r
a
tio
n
s
1
a
n
d
3
,
in
d
icate
d
b
y
n
e
g
ativ
e
R²
s
co
r
es
,
alth
o
u
g
h
it
a
ch
iev
ed
r
ea
s
o
n
ab
le
m
etr
ics
in
c
o
n
f
ig
u
r
atio
n
2
.
T
h
e
p
r
o
p
o
s
ed
STGP
-
Net
(
1
D
-
C
NN+
L
STM
)
m
o
d
el
co
n
s
is
ten
tly
d
em
o
n
s
tr
ated
s
tr
o
n
g
p
er
f
o
r
m
an
ce
;
it
ac
h
iev
ed
th
e
lo
west
MA
E
/R
MSE
an
d
h
ig
h
est
R²
in
c
o
n
f
ig
u
r
atio
n
2
,
th
e
h
ig
h
est
R²
i
n
c
o
n
f
ig
u
r
ati
o
n
3
(
with
MA
E
/R
MSE
v
alu
es
v
er
y
clo
s
e
to
th
e
b
est),
an
d
c
o
m
p
etitiv
e
r
esu
lts
in
co
n
f
i
g
u
r
atio
n
1
.
B
ased
o
n
th
is
co
m
p
r
eh
en
s
iv
e
ev
alu
atio
n
ac
r
o
s
s
m
u
ltip
le
m
etr
ics
an
d
co
n
f
ig
u
r
atio
n
s
,
th
e
STGP
-
Net
m
o
d
el
em
er
g
es
as
th
e
m
o
s
t
r
o
b
u
s
t
an
d
ef
f
ec
tiv
e
p
r
e
d
icto
r
,
d
is
p
l
ay
in
g
s
u
p
er
io
r
o
v
er
all
p
er
f
o
r
m
an
ce
an
d
g
r
ea
ter
co
n
s
is
ten
cy
co
m
p
ar
e
d
to
th
e
1
D
-
C
NN+
R
NN
an
d
1
D
-
C
NN+
B
iLST
M
h
y
b
r
id
m
o
d
els.
T
o
f
u
r
th
e
r
a
s
s
e
s
s
t
h
e
p
r
ed
i
c
ti
v
e
c
ap
a
b
i
li
t
i
e
s
,
F
i
g
u
r
e
s
6
to
8
p
l
o
t
th
e
f
o
r
e
c
a
s
t
ed
v
a
lu
e
s
f
r
o
m
th
e
1D
-
C
N
N
+
R
N
N
,
S
T
G
P
-
N
e
t
,
a
n
d
1
D
-
C
N
N
+
B
i
L
S
T
M
m
o
d
e
l
s
a
l
o
n
g
s
i
d
e
t
h
e
t
r
u
e
h
i
s
t
o
r
ic
a
l
g
o
ld
p
r
i
c
e
s
f
o
r
e
a
c
h
o
f
th
e
t
h
r
e
e
co
n
f
ig
u
r
at
i
o
n
s
.
T
h
e
v
i
s
u
a
l
r
e
s
u
l
t
s
p
r
e
s
e
n
t
e
d
in
F
ig
u
r
e
s
6
to
8
f
u
r
t
h
e
r
s
u
p
p
o
r
t
t
h
e
q
u
an
t
i
t
a
t
i
v
e
f
in
d
in
g
s
.
W
h
i
l
e
a
l
l
m
o
d
e
l
s
a
t
t
e
m
p
t
to
c
ap
t
u
r
e
t
h
e
t
r
en
d
o
f
th
e
g
o
l
d
p
r
ic
e,
t
h
e
p
l
o
t
s
v
i
s
u
a
l
ly
c
o
n
f
i
r
m
t
h
e
i
n
co
n
s
i
s
t
en
c
i
e
s
o
b
s
er
v
ed
in
t
h
e
m
e
t
r
i
c
s
f
o
r
th
e
1
D
-
C
N
N
+
R
N
N
a
n
d
1
D
-
C
N
N
+
B
i
L
S
T
M
m
o
d
e
l
s
,
p
a
r
t
ic
u
l
ar
l
y
ev
i
d
en
t
i
n
c
o
n
f
i
g
u
r
a
t
io
n
s
2
a
n
d
3
wh
e
r
e
th
e
i
r
p
r
ed
i
c
t
i
o
n
s
s
o
m
e
t
i
m
e
s
d
ev
i
a
t
e
s
i
g
n
i
f
i
ca
n
t
ly
f
r
o
m
th
e
t
r
u
e
v
al
u
e
s
.
T
h
e
S
T
G
P
-
N
e
t
(
1
D
-
C
NN
+
L
S
T
M
)
m
o
d
e
l
's
p
r
ed
i
c
t
io
n
s
g
e
n
er
a
l
l
y
ap
p
e
a
r
t
o
t
r
ac
k
t
h
e
a
c
t
u
a
l
p
r
i
c
e
m
o
v
e
m
en
t
s
m
o
r
e
c
l
o
s
e
ly
a
cr
o
s
s
t
h
e
c
o
n
f
ig
u
r
a
t
i
o
n
s
,
a
l
ig
n
in
g
w
i
t
h
i
t
s
s
tr
o
n
g
er
p
e
r
f
o
r
m
a
n
c
e
m
e
t
r
i
c
s
.
Fig
u
r
e
6
.
T
r
u
e
v
al
u
es v
s
.
p
r
ed
i
cted
v
alu
es o
f
c
o
n
f
ig
u
r
atio
n
1
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Hyb
r
id
d
ee
p
lea
r
n
in
g
mo
d
el
f
o
r
en
h
a
n
ce
d
s
h
o
r
t
-
term g
o
ld
p
r
ice
…
(
Ho
a
n
g
Ha
N
g
u
ye
n
)
3237
Fig
u
r
e
7
.
T
r
u
e
v
al
u
es v
s
.
p
r
ed
i
cted
v
alu
es o
f
c
o
n
f
ig
u
r
atio
n
2
Fig
u
r
e
8
.
T
r
u
e
v
al
u
es v
s
.
p
r
ed
i
cted
v
alu
es o
f
c
o
n
f
ig
u
r
atio
n
3
I
n
s
u
m
m
ar
y
,
th
is
ex
p
er
im
e
n
tal
s
ec
tio
n
d
etailed
th
e
d
ata
p
r
ep
ar
atio
n
u
s
in
g
s
lid
in
g
win
d
o
ws,
th
e
ev
alu
atio
n
f
r
am
ewo
r
k
in
clu
d
i
n
g
m
etr
ics
a
n
d
lo
s
s
an
al
y
s
is
,
an
d
th
e
co
m
p
ar
ativ
e
p
er
f
o
r
m
an
ce
ass
ess
m
en
t
o
f
th
r
ee
h
y
b
r
i
d
m
o
d
els.
T
h
e
co
m
p
r
eh
e
n
s
iv
e
r
esu
lts
co
n
s
is
ten
tly
in
d
icate
th
at
th
e
p
r
o
p
o
s
ed
STGP
-
Net
m
o
d
el
p
r
o
v
id
es
e
n
h
an
ce
d
an
d
m
o
r
e
r
eliab
le
s
h
o
r
t
-
ter
m
g
o
ld
p
r
ice
f
o
r
ec
asti
n
g
u
n
d
er
t
h
e
test
ed
co
n
d
itio
n
s
.
C
o
m
p
ar
ed
to
t
h
e
1
D
-
C
NN
h
y
b
r
i
d
m
o
d
els
u
tili
zin
g
s
tan
d
ar
d
R
NN
o
r
B
iLST
M
lay
er
s
,
STGP
-
Net
d
em
o
n
s
tr
ated
s
u
p
e
r
io
r
p
er
f
o
r
m
an
ce
.
5.
CO
NCLU
SI
O
N
T
h
is
s
tu
d
y
in
v
esti
g
ated
h
y
b
r
id
d
ee
p
lear
n
in
g
m
o
d
els
f
o
r
s
h
o
r
t
-
ter
m
g
o
ld
p
r
ice
f
o
r
ec
asti
n
g
,
p
r
o
p
o
s
in
g
STGP
-
Net
(
a
1
D
-
C
NN
-
L
STM
ar
ch
itectu
r
e
)
an
d
c
o
m
p
a
r
in
g
it
ag
ain
s
t
1
D
-
C
NN
co
m
b
in
ed
with
s
tan
d
ar
d
R
NN
an
d
B
iLST
M
ac
r
o
s
s
v
ar
io
u
s
win
d
o
w
co
n
f
ig
u
r
atio
n
s
.
C
o
m
p
r
eh
en
s
iv
e
ex
p
er
im
en
tal
r
esu
lts
co
n
s
is
ten
tly
s
h
o
wed
th
at
STGP
-
Net
o
u
tp
e
r
f
o
r
m
e
d
th
e
b
aselin
e
m
o
d
els
in
ter
m
s
o
f
p
r
ed
ictiv
e
ac
cu
r
a
cy
an
d
r
o
b
u
s
tn
ess
b
ased
o
n
MA
E
,
R
MSE
,
an
d
R²
m
etr
ics.
T
h
ese
f
in
d
in
g
s
estab
lis
h
th
e
p
r
o
p
o
s
ed
1
D
-
C
NN
-
L
STM
co
m
b
in
atio
n
as
an
ef
f
ec
tiv
e
ap
p
r
o
ac
h
f
o
r
th
is
ch
allen
g
in
g
f
o
r
ec
asti
n
g
t
ask
.
Fu
tu
r
e
r
esear
ch
c
o
u
ld
e
x
ten
d
th
is
wo
r
k
b
y
in
co
r
p
o
r
atin
g
ad
d
itio
n
al
f
ea
t
u
r
es
lik
e
m
ac
r
o
ec
o
n
o
m
ic
in
d
icato
r
s
o
r
m
ar
k
et
s
en
tim
en
t,
ex
p
lo
r
in
g
m
o
r
e
ad
v
an
ce
d
h
y
p
er
p
ar
a
m
eter
o
p
ti
m
izatio
n
m
eth
o
d
s
,
o
r
a
d
ap
tin
g
th
e
m
o
d
el
f
o
r
l
o
n
g
er
-
ter
m
p
r
e
d
ictio
n
h
o
r
izo
n
s
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
T
h
is
s
tu
d
y
was f
u
n
d
e
d
b
y
Hu
e
Un
iv
er
s
ity
,
Vietn
am
(
g
r
an
t n
u
m
b
er
DHH2
0
2
5
-
01
-
2
2
5
)
.
Evaluation Warning : The document was created with Spire.PDF for Python.