Co
m
pu
t
er
Science
a
nd
I
nfo
r
m
a
t
io
n T
ec
hn
o
lo
g
ies
Vo
l.
7
,
No
.
3
,
No
v
em
b
er
20
26
,
p
p
.
291
~
303
I
SS
N:
2722
-
3
2
2
1
,
DOI
:
1
0
.
1
1
5
9
1
/csi
t
.
v
7
i
3
.
p
2
9
1
-
3
0
3
291
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ia
e
s
p
r
ime.
co
m/in
d
ex
.
p
h
p
/csi
t
Tim
e series
forec
a
sting
:
a
com
pa
ra
tive a
na
ly
sis
of A
RIM
A,
LST
M
,
a
nd TFT
mo
dels with
miss
i
ng
data
hand
ling
M
a
ry
a
m
H
o
s
s
eini
,
M
o
ha
m
a
d F
o
ro
uza
nfa
r
D
e
p
a
r
t
me
n
t
o
f
S
y
s
t
e
ms
En
g
i
n
e
e
r
i
n
g
,
Éc
o
l
e
d
e
T
e
c
h
n
o
l
o
g
i
e
S
u
p
é
r
i
e
u
r
e
(
ÉT
S
)
,
U
n
i
v
e
r
s
i
t
é
d
u
Q
u
é
b
e
c
,
M
o
n
t
r
e
a
l
,
C
a
n
a
d
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Mar
13
,
2
0
2
6
R
ev
is
ed
May
17
,
2
0
2
6
Acc
ep
ted
J
u
n
8
,
2
0
2
6
Ti
m
e
se
ries
fo
re
c
a
stin
g
p
lay
s
a
c
rit
ica
l
ro
le
in
fin
a
n
c
e
,
h
e
a
lt
h
c
a
re
,
a
n
d
e
n
e
rg
y
a
p
p
li
c
a
ti
o
n
s,
w
h
e
re
a
c
c
u
ra
te p
re
d
ictio
n
s su
p
p
o
rt
d
e
c
isi
o
n
-
m
a
k
in
g
a
n
d
o
p
e
ra
ti
o
n
a
l
e
fficie
n
c
y
.
Trad
it
io
n
a
l
a
p
p
ro
a
c
h
e
s
su
c
h
a
s
a
u
t
o
r
e
g
re
ss
iv
e
in
teg
ra
ted
m
o
v
i
n
g
a
v
e
ra
g
e
(ARI
M
A)
p
e
rfo
rm
we
ll
fo
r
li
n
e
a
r
p
a
tt
e
rn
s
b
u
t
o
ften
stru
g
g
le
with
n
o
n
li
n
e
a
r
a
n
d
c
o
m
p
lex
tem
p
o
ra
l
d
e
p
e
n
d
e
n
c
ie
s
fo
u
n
d
in
re
a
l
-
wo
rld
d
a
ta.
Alth
o
u
g
h
d
e
e
p
lea
rn
in
g
m
e
th
o
d
s
su
c
h
a
s
lo
n
g
sh
o
rt
-
term
m
e
m
o
ry
(LS
TM
)
n
e
two
rk
s
a
n
d
Tran
sfo
rm
e
r
-
b
a
se
d
m
o
d
e
ls
h
a
v
e
sh
o
wn
p
ro
m
ise
,
c
o
m
p
re
h
e
n
siv
e
e
v
a
lu
a
t
io
n
s
a
c
ro
ss
m
u
lt
i
p
le
d
o
m
a
in
s
a
n
d
u
n
d
e
r
m
issin
g
-
d
a
ta
c
o
n
d
it
i
o
n
s
re
m
a
in
li
m
it
e
d
.
Th
is
st
u
d
y
c
o
m
p
a
re
s
ARIMA,
LS
TM
,
a
n
d
tem
p
o
ra
l
f
u
sio
n
tran
sfo
rm
e
r
(TF
T)
m
o
d
e
ls
a
c
ro
ss
th
re
e
a
p
p
li
c
a
ti
o
n
s:
st
o
c
k
p
rice
fo
re
c
a
stin
g
u
sin
g
S
&
P
5
0
0
d
a
ta,
h
e
a
rt
-
ra
te
p
re
d
ictio
n
fro
m
e
lec
tro
c
a
rd
i
o
g
ra
m
(
ECG
)
-
d
e
riv
e
d
si
g
n
a
ls,
a
n
d
e
lec
tri
c
it
y
d
e
m
a
n
d
fo
re
c
a
stin
g
u
si
n
g
P
e
n
n
s
y
lv
a
n
ia
–
Ne
w Je
rse
y
–
M
a
ry
lan
d
(
P
J
M
)
p
o
we
r
g
rid
d
a
ta.
T
o
e
v
a
l
u
a
te
ro
b
u
st
n
e
ss
u
n
d
e
r
re
a
li
stic
c
o
n
d
it
i
o
n
s,
v
a
ry
i
n
g
lev
e
ls o
f
m
issin
g
n
e
ss
we
re
in
tr
o
d
u
c
e
d
u
si
n
g
m
issi
n
g
c
o
m
p
lete
ly
a
t
ra
n
d
o
m
(M
CAR)
a
n
d
m
issin
g
a
t
ra
n
d
o
m
(
M
AR)
m
e
c
h
a
n
ism
s.
M
issin
g
v
a
l
u
e
s
we
re
h
a
n
d
led
u
sin
g
f
o
rwa
rd
fil
l,
li
n
e
a
r
i
n
te
rp
o
latio
n
,
a
n
d
k
-
n
e
a
re
st
n
e
i
g
h
b
o
r
s
(k
-
NN
)
imp
u
tati
o
n
.
Re
su
lt
s
sh
o
w
th
a
t
TF
T
c
o
n
siste
n
t
ly
a
c
h
iev
e
d
su
p
e
rio
r
fo
re
c
a
stin
g
p
e
rfo
rm
a
n
c
e
a
n
d
d
e
m
o
n
stra
ted
g
re
a
ter
ro
b
u
stn
e
ss
to
in
c
re
a
sin
g
m
issin
g
n
e
ss
,
w
h
il
e
k
-
NN
g
e
n
e
ra
ll
y
p
ro
v
i
d
e
d
t
h
e
m
o
st
e
ffe
c
ti
v
e
i
m
p
u
tatio
n
pe
rfo
rm
a
n
c
e
a
c
ro
ss
d
a
tas
e
ts.
K
ey
w
o
r
d
s
:
AR
I
MA
L
o
n
g
s
h
o
r
t
-
ter
m
m
e
m
o
r
y
Miss
in
g
d
ata
im
p
u
tatio
n
T
em
p
o
r
al
f
u
s
io
n
tr
an
s
f
o
r
m
er
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
:
Ma
r
y
am
Ho
s
s
ein
i
Dep
ar
tm
en
t o
f
Sy
s
tem
s
E
n
g
in
ee
r
in
g
,
É
c
o
le
d
e
T
ec
h
n
o
lo
g
ie
Su
p
ér
ieu
r
e
(
É
T
S)
,
Un
iv
er
s
ité
d
u
Qu
éb
ec
Mo
n
tr
ea
l,
C
an
ad
a
E
m
ail:
m
ar
y
am
y
2
k
8
7
6
@
g
m
ai
l.c
o
m
1.
I
NT
RO
D
UCT
I
O
N
T
im
e
-
s
er
ies
f
o
r
ec
asti
n
g
h
as
b
ec
o
m
e
in
c
r
ea
s
in
g
ly
im
p
o
r
tan
t
in
d
ata
-
d
r
i
v
en
a
p
p
licatio
n
s
;
h
o
wev
er
,
ac
h
iev
in
g
r
eliab
le
p
r
ed
ictio
n
s
in
r
ea
l
-
wo
r
ld
s
ettin
g
s
r
em
ai
n
s
ch
allen
g
in
g
b
ec
a
u
s
e
tem
p
o
r
al
d
ata
f
r
eq
u
en
tly
ex
h
ib
it
n
o
n
lin
ea
r
ity
,
lo
n
g
-
r
an
g
e
d
ep
e
n
d
en
cies,
n
o
is
e,
an
d
i
n
co
m
p
lete
o
b
s
er
v
atio
n
s
.
I
n
f
i
n
an
ce
,
th
e
ab
ilit
y
to
p
r
ed
ict
s
to
ck
p
r
ices,
m
ar
k
et
tr
en
d
s
,
an
d
ec
o
n
o
m
ic
in
d
icat
o
r
s
is
cr
u
cial
f
o
r
r
is
k
m
an
ag
em
en
t,
in
v
estme
n
t
s
tr
ateg
ies,
an
d
f
in
an
cial
p
lan
n
in
g
[1
]
,
[
2]
.
Similar
ly
,
in
h
ea
lth
ca
r
e,
f
o
r
ec
asti
n
g
v
ital
s
ig
n
s
lik
e
h
ea
r
t
r
ate
d
u
r
in
g
s
leep
ca
n
e
n
ab
le
ea
r
ly
d
etec
tio
n
o
f
ab
n
o
r
m
alities
s
u
ch
as
s
leep
d
is
o
r
d
er
s
,
im
p
r
o
v
in
g
p
atien
t
o
u
tco
m
es
[
3
]
. P
r
ed
ictin
g
tu
m
o
r
g
r
o
wth
u
s
in
g
im
ag
in
g
b
io
m
ar
k
er
s
ca
n
g
u
id
e
o
n
c
o
lo
g
is
ts
in
ass
e
s
s
in
g
ca
n
ce
r
p
r
o
g
r
es
s
io
n
an
d
ad
ju
s
tin
g
tr
ea
tm
en
t
p
lan
s
p
r
o
ac
tiv
ely
[
4
]
.
Ad
d
itio
n
ally
,
f
o
r
ec
asti
n
g
p
atien
t
r
esp
o
n
s
es
to
ch
em
o
th
er
ap
y
b
ased
o
n
p
r
io
r
t
r
ea
tm
en
ts
an
d
g
en
o
m
ic
d
ata
ca
n
h
elp
p
e
r
s
o
n
alize
th
er
ap
y
r
eg
im
en
s
,
o
p
ti
m
izin
g
ef
f
ec
tiv
e
n
ess
an
d
r
ed
u
cin
g
s
id
e
ef
f
ec
ts
.
I
n
th
e
en
er
g
y
s
ec
to
r
,
ac
c
u
r
ate
p
r
ed
ictio
n
s
o
f
elec
tr
icity
d
e
m
an
d
ar
e
cr
itical
f
o
r
o
p
tim
izin
g
p
o
wer
g
r
id
o
p
er
atio
n
s
an
d
in
te
g
r
atin
g
r
en
ewa
b
le
en
er
g
y
s
o
u
r
ce
s
in
to
th
e
g
r
id
[5
]
–
[
7]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
7
2
2
-
3
2
2
1
C
o
m
p
u
t Sci
I
n
f
T
ec
h
n
o
l
,
Vo
l.
7
,
No
.
3
,
No
v
em
b
er
20
26
:
2
9
1
-
303
292
T
h
e
c
h
alle
n
g
e
o
f
f
o
r
ec
a
s
ti
n
g
ti
m
e
s
er
ies
d
a
ta
ar
is
es
f
r
o
m
its
o
f
te
n
v
o
la
tile
a
n
d
c
o
m
p
le
x
n
a
t
u
r
e,
w
h
e
r
e
n
o
n
-
li
n
ea
r
p
a
tte
r
n
s
,
n
o
is
e
,
an
d
lo
n
g
-
te
r
m
d
e
p
e
n
d
e
n
c
ies
m
ak
e
ac
c
u
r
a
te
p
r
e
d
i
cti
o
n
s
d
i
f
f
i
cu
l
t.
F
u
r
th
er
m
o
r
e,
r
e
al
-
wo
r
l
d
ti
m
e
s
er
ies
d
at
a
o
f
ten
s
u
f
f
er
f
r
o
m
m
is
s
i
n
g
v
a
lu
e
s
,
wh
ic
h
ca
n
a
r
is
e
f
r
o
m
v
a
r
i
o
u
s
s
o
u
r
c
es
s
u
ch
as
s
en
s
o
r
f
ail
u
r
es
,
d
at
a
en
tr
y
er
r
o
r
s
,
o
r
i
r
r
e
g
u
la
r
s
am
p
li
n
g
.
Miss
in
g
d
at
a
ca
n
s
i
g
n
i
f
ic
an
tl
y
i
m
p
a
ct
t
h
e
ac
c
u
r
ac
y
an
d
r
el
ia
b
il
it
y
o
f
ti
m
e
s
e
r
i
es
f
o
r
e
c
asti
n
g
m
o
d
e
ls
,
le
ad
in
g
t
o
b
i
as
ed
o
r
in
c
o
m
p
l
ete
p
r
ed
ict
io
n
s
.
T
h
er
ef
o
r
e
,
ef
f
e
cti
v
e
h
a
n
d
li
n
g
o
f
m
is
s
i
n
g
d
ata
is
c
r
u
cial
t
o
e
n
s
u
r
e
t
h
e
r
o
b
u
s
t
n
ess
a
n
d
r
el
ia
b
il
it
y
o
f
f
o
r
ec
asti
n
g
m
o
d
els
.
His
to
r
ically
,
th
e
a
u
to
r
eg
r
ess
iv
e
in
teg
r
ated
m
o
v
in
g
a
v
er
ag
e
(
AR
I
MA
)
m
o
d
el
h
as
b
ee
n
a
d
o
m
i
n
an
t
ap
p
r
o
ac
h
in
tim
e
s
er
ies
f
o
r
ec
a
s
tin
g
[
8
]
.
AR
I
MA
is
v
alu
ed
f
o
r
its
s
im
p
licity
,
in
ter
p
r
etab
ilit
y
,
an
d
ef
f
icien
cy
in
h
an
d
lin
g
lin
ea
r
r
elatio
n
s
h
ip
s
.
I
t
is
p
ar
ticu
la
r
ly
a
d
ep
t
at
f
o
r
ec
asti
n
g
f
in
an
cial
tim
e
s
er
ie
s
th
at
ca
n
b
e
m
a
d
e
s
tatio
n
ar
y
th
r
o
u
g
h
d
i
f
f
er
en
ci
n
g
,
an
d
it
h
as
also
b
ee
n
ap
p
lied
in
en
er
g
y
d
em
a
n
d
f
o
r
ec
asti
n
g
.
Ho
wev
er
,
AR
I
MA
m
o
d
els
h
av
e
n
o
tab
le
lim
itatio
n
s
,
p
ar
ticu
la
r
ly
in
th
ei
r
in
ab
ilit
y
to
ca
p
tu
r
e
n
o
n
-
lin
ea
r
r
elatio
n
s
h
ip
s
an
d
co
m
p
lex
tem
p
o
r
al
d
e
p
en
d
e
n
cies.
T
h
ese
lim
itatio
n
s
h
av
e
p
r
o
m
p
ted
th
e
d
ev
elo
p
m
en
t
o
f
m
o
r
e
ad
v
an
ce
d
m
o
d
els s
u
ch
as
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
L
STM
)
n
etwo
r
k
s
an
d
T
r
a
n
s
f
o
r
m
er
m
o
d
els
[
9
]
.
Dee
p
lear
n
in
g
h
as
r
ev
o
lu
tio
n
ized
th
e
f
ield
o
f
s
eq
u
e
n
tial
d
ata
p
r
o
ce
s
s
in
g
,
o
f
f
er
i
n
g
p
o
wer
f
u
l
ar
ch
itectu
r
es
ca
p
ab
le
o
f
ca
p
t
u
r
in
g
c
o
m
p
lex
p
atter
n
s
in
d
iv
er
s
e
ap
p
licatio
n
s
[
1
0
]
–
[
1
4
]
.
On
e
o
f
th
e
k
ey
ad
v
an
ce
m
e
n
ts
in
d
ee
p
lear
n
in
g
f
o
r
tim
e
s
er
ies
f
o
r
ec
asti
n
g
is
th
e
u
s
e
o
f
r
ec
u
r
r
en
t
n
eu
r
al
n
etwo
r
k
s
(
R
NNs),
p
ar
ticu
lar
ly
L
STM
n
etwo
r
k
s
.
L
STM
s
h
av
e
em
e
r
g
ed
as
a
p
o
wer
f
u
l
to
o
l
f
o
r
m
o
d
elin
g
l
o
n
g
-
ter
m
d
e
p
en
d
e
n
cies
an
d
ca
p
tu
r
in
g
n
o
n
-
lin
ea
r
p
att
er
n
s
in
s
eq
u
en
tial
d
ata
[
1
5
]
,
[
16]
.
T
h
e
y
ar
e
esp
ec
ially
ef
f
ec
tiv
e
in
ap
p
licatio
n
s
s
u
ch
as
f
in
a
n
cial
f
o
r
ec
asti
n
g
,
wh
er
e
p
ast
ev
e
n
ts
ca
n
h
av
e
d
elay
ed
ef
f
ec
ts
o
n
f
u
t
u
r
e
m
ar
k
et
b
eh
av
io
r
s
,
h
ea
lth
ca
r
e
f
o
r
p
r
ed
ictin
g
v
ital sig
n
s
,
an
d
en
er
g
y
f
o
r
ec
asti
n
g
f
o
r
m
o
d
elin
g
co
n
s
u
m
p
tio
n
p
at
ter
n
s
in
f
lu
en
ce
d
b
y
f
ac
to
r
s
lik
e
wea
t
h
e
r
a
n
d
h
u
m
a
n
ac
tiv
ity
.
Ho
wev
er
,
d
esp
ite
t
h
ese
ad
v
an
ta
g
es,
L
STM
s
s
tr
u
g
g
le
with
ca
p
tu
r
in
g
v
er
y
lo
n
g
-
r
a
n
g
e
d
ep
en
d
en
cie
s
d
u
e
to
th
eir
s
eq
u
en
tial
n
atu
r
e,
wh
ich
ca
n
lead
to
v
an
is
h
i
n
g
g
r
a
d
ien
t
is
s
u
es.
T
o
ad
d
r
ess
th
ese
s
h
o
r
tco
m
i
n
g
s
,
T
r
an
s
f
o
r
m
er
m
o
d
els
h
av
e
em
er
g
e
d
as
a
s
tate
-
of
-
th
e
-
ar
t
alter
n
ativ
e,
p
ar
ticu
lar
ly
ex
ce
llin
g
in
n
atu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
an
d
in
cr
ea
s
in
g
ly
b
ein
g
ap
p
lied
to
tim
e
s
er
ies
f
o
r
ec
asti
n
g
[
1
7
]
.
Un
lik
e
L
STM
s
,
T
r
an
s
f
o
r
m
er
s
u
s
e
s
elf
-
atten
tio
n
m
ec
h
an
is
m
s
to
p
r
o
ce
s
s
en
tire
s
e
q
u
en
ce
s
in
p
ar
allel,
s
ig
n
if
ican
tly
im
p
r
o
v
in
g
c
o
m
p
u
tatio
n
al
ef
f
icien
cy
an
d
e
n
ab
lin
g
th
e
m
o
d
elin
g
o
f
m
u
ch
lo
n
g
er
-
r
an
g
e
d
ep
en
d
e
n
cies
[
1
8
]
,
[
19]
.
T
h
is
ab
ilit
y
to
ca
p
tu
r
e
co
m
p
le
x
te
m
p
o
r
al
p
atter
n
s
wh
ile
s
ca
lin
g
ef
f
ec
tiv
ely
m
ak
es
T
r
an
s
f
o
r
m
e
r
s
h
ig
h
ly
p
r
o
m
is
in
g
f
o
r
a
p
p
licatio
n
s
in
f
in
an
cial,
h
ea
lth
ca
r
e,
an
d
en
er
g
y
f
o
r
ec
as
tin
g
.
Desp
ite
s
ig
n
if
ican
t
ad
v
an
ce
s
in
tim
e
-
s
er
ies
f
o
r
ec
asti
n
g
,
s
ev
er
al
lim
itatio
n
s
r
em
ain
in
th
e
ex
is
tin
g
liter
atu
r
e.
Pre
v
io
u
s
s
tu
d
ies
h
a
v
e
o
f
ten
f
o
cu
s
ed
o
n
ev
al
u
atin
g
f
o
r
ec
asti
n
g
m
o
d
els
with
in
a
s
in
g
le
a
p
p
licatio
n
d
o
m
ain
,
s
u
ch
as
s
to
ck
m
ar
k
ets,
h
ea
lth
ca
r
e
m
o
n
it
o
r
in
g
,
o
r
e
n
er
g
y
s
y
s
tem
s
s
ep
ar
atel
y
.
Mo
r
eo
v
e
r
,
m
o
s
t
co
m
p
ar
ativ
e
s
tu
d
ies
em
p
h
asiz
e
p
r
ed
ictio
n
ac
cu
r
ac
y
u
n
d
er
c
o
m
p
lete
d
atasets
an
d
p
ay
lim
ited
atten
tio
n
to
th
e
im
p
ac
t
o
f
m
is
s
in
g
d
ata,
d
esp
ite
m
is
s
in
g
n
ess
b
ein
g
a
co
m
m
o
n
is
s
u
e
in
r
ea
l
-
wo
r
ld
en
v
ir
o
n
m
en
ts
d
u
e
t
o
s
en
s
o
r
f
ailu
r
es,
co
m
m
u
n
icatio
n
er
r
o
r
s
,
ir
r
eg
u
lar
s
am
p
lin
g
,
an
d
in
co
m
p
lete
r
ec
o
r
d
s
.
E
x
is
tin
g
r
esea
r
ch
r
ar
ely
p
r
o
v
id
es
a
u
n
if
ie
d
ev
al
u
atio
n
o
f
tr
ad
iti
o
n
al
s
tatis
tical
m
o
d
els
an
d
r
e
ce
n
t
d
ee
p
lear
n
i
n
g
ar
ch
itectu
r
es
ac
r
o
s
s
m
u
ltip
le
d
o
m
ain
s
u
n
d
e
r
co
n
tr
o
lled
m
i
s
s
in
g
-
d
ata
s
ce
n
ar
io
s
.
I
n
p
ar
ticu
lar
,
co
m
p
ar
ativ
e
in
v
esti
g
atio
n
s
o
f
AR
I
MA
,
L
STM
,
an
d
tem
p
o
r
al
f
u
s
io
n
tr
an
s
f
o
r
m
er
(
T
FT)
u
n
d
er
d
if
f
er
en
t
m
is
s
in
g
-
d
ata
m
ec
h
a
n
is
m
s
an
d
im
p
u
tatio
n
s
tr
ateg
ies
r
em
ain
lim
ited
.
T
h
i
s
g
ap
r
estricts
u
n
d
er
s
tan
d
in
g
o
f
m
o
d
el
r
o
b
u
s
tn
ess
an
d
p
r
ac
t
ical
ap
p
licab
ilit
y
in
r
ea
l
-
wo
r
ld
f
o
r
ec
asti
n
g
s
ettin
g
s
.
T
o
ad
d
r
ess
th
ese
lim
itatio
n
s
,
th
is
s
tu
d
y
p
r
esen
ts
a
co
m
p
r
eh
e
n
s
iv
e
co
m
p
a
r
ativ
e
an
al
y
s
is
o
f
AR
I
MA
,
L
STM
,
an
d
T
FT
m
o
d
els
ac
r
o
s
s
th
r
ee
r
ep
r
esen
tativ
e
d
o
m
ain
s
:
f
in
an
ce
,
h
ea
lth
ca
r
e
,
an
d
en
e
r
g
y
.
T
h
e
n
o
v
elty
o
f
th
is
wo
r
k
lies
in
:
i
)
e
v
alu
atin
g
m
o
d
el
p
e
r
f
o
r
m
an
ce
ac
r
o
s
s
m
u
ltip
le
ap
p
licatio
n
d
o
m
ain
s
r
ath
er
th
a
n
a
s
in
g
le
d
ataset;
ii
)
s
y
s
tem
atica
lly
in
v
esti
g
atin
g
m
o
d
el
r
o
b
u
s
tn
ess
u
n
d
er
r
ea
lis
tic
m
is
s
in
g
-
d
ata
co
n
d
itio
n
s
u
s
in
g
b
o
t
h
m
is
s
in
g
co
m
p
letely
at
r
an
d
o
m
(
MCAR
)
an
d
m
is
s
in
g
at
r
an
d
o
m
(
MA
R
)
m
ec
h
an
is
m
s
;
iii
)
co
m
p
ar
in
g
m
u
ltip
le
im
p
u
tatio
n
tech
n
iq
u
es,
in
clu
d
in
g
f
o
r
war
d
f
ill,
lin
ea
r
in
te
r
p
o
latio
n
,
an
d
k
-
n
ea
r
est
n
eig
h
b
o
r
s
(
k
-
NN
)
;
an
d
iv
)
an
aly
zin
g
th
e
ca
p
a
b
ilit
y
o
f
ad
v
an
ce
d
tr
an
s
f
o
r
m
er
-
b
ased
ar
ch
itectu
r
es
to
m
ain
tain
f
o
r
ec
asti
n
g
p
er
f
o
r
m
an
ce
u
n
d
er
in
cr
ea
s
in
g
m
is
s
in
g
n
ess
.
T
h
ese
co
n
tr
i
b
u
tio
n
s
p
r
o
v
id
e
p
r
ac
tical
i
n
s
ig
h
ts
in
to
m
o
d
el
s
elec
tio
n
a
n
d
m
is
s
in
g
-
d
ata
h
an
d
lin
g
s
tr
ateg
ies f
o
r
r
ea
l
-
wo
r
l
d
tim
e
-
s
er
ies f
o
r
ec
asti
n
g
ap
p
licatio
n
s
.
2.
M
E
T
H
O
D
2
.
1
.
Da
t
a
s
et
s
W
e
u
s
ed
th
r
ee
d
atasets
ac
r
o
s
s
d
if
f
er
en
t
d
o
m
ain
s
: f
in
a
n
ce
,
h
e
alth
ca
r
e,
an
d
e
n
er
g
y
.
2
.
1
.
1
.
F
ina
nce
da
t
a
s
et
T
h
e
f
in
an
cial
d
ataset
u
s
ed
in
th
is
s
tu
d
y
co
n
s
is
t
s
o
f
d
aily
s
to
ck
p
r
ices
f
r
o
m
th
e
S&
P
5
0
0
in
d
ex
.
T
h
is
d
ataset
co
v
er
s
a
1
0
-
y
ea
r
p
er
io
d
,
p
r
o
v
id
i
n
g
a
r
ich
h
is
to
r
ical
co
n
te
x
t
f
o
r
f
o
r
ec
asti
n
g
s
to
ck
p
r
ice
tr
en
d
s
.
T
h
e
d
ataset
in
clu
d
es th
e
f
o
llo
win
g
v
ar
iab
les:
−
Op
en
p
r
ice:
th
e
p
r
ice
at
wh
ich
th
e
s
to
ck
m
ar
k
et
o
p
e
n
ed
f
o
r
e
ac
h
tr
ad
in
g
d
ay
.
Evaluation Warning : The document was created with Spire.PDF for Python.
C
o
m
p
u
t Sci
I
n
f
T
ec
h
n
o
l
I
SS
N:
2722
-
3
2
2
1
Time
s
er
ies fo
r
ec
a
s
tin
g
:
a
co
mp
a
r
a
tive
a
n
a
lysi
s
o
f A
R
I
MA,
LS
TM
,
a
n
d
TFT m
o
d
els
…
(
Ma
r
ya
m
Ho
s
s
ein
i
)
293
−
C
lo
s
e
p
r
ice:
th
e
f
in
al
p
r
ice
at
wh
ich
th
e
s
to
ck
m
a
r
k
et
clo
s
ed
f
o
r
ea
c
h
tr
ad
in
g
d
ay
.
−
Hig
h
p
r
ice:
th
e
h
ig
h
est p
r
ice
r
ea
ch
ed
d
u
r
in
g
t
h
e
tr
ad
in
g
d
a
y
.
−
L
o
w
p
r
ice:
th
e
lo
west p
r
ice
r
e
ac
h
ed
d
u
r
in
g
t
h
e
tr
ad
in
g
d
a
y
.
−
Vo
lu
m
e
: th
e
to
tal
n
u
m
b
er
o
f
s
h
ar
es tr
ad
ed
d
u
r
in
g
th
e
tr
a
d
in
g
d
ay
.
Ou
r
g
o
al
o
n
t
h
is
d
ataset
was
to
p
r
e
d
ict
th
e
clo
s
in
g
p
r
ice
o
f
t
h
e
S&
P
5
0
0
in
d
e
x
f
o
r
th
e
n
ex
t
d
ay
,
b
ased
o
n
th
e
h
is
to
r
ical
v
alu
es o
f
th
e
a
f
o
r
em
en
tio
n
ed
v
ar
iab
les.
2
.
1
.
2
.
H
ea
lt
hca
re
da
t
a
s
et
T
h
e
h
ea
lth
ca
r
e
d
ataset,
d
er
iv
e
d
f
r
o
m
th
e
Ph
y
s
io
n
et
Ap
n
ea
-
E
C
G
Data
b
ase,
co
m
p
r
is
es
h
ea
r
t
r
ate
d
ata
d
er
iv
ed
f
r
o
m
elec
tr
o
ca
r
d
io
g
r
am
(
E
C
G
)
s
ig
n
als
f
o
r
3
5
in
d
iv
id
u
als
d
u
r
in
g
s
leep
,
w
h
ich
co
n
tain
s
an
n
o
tated
E
C
G
r
ec
o
r
d
in
g
s
u
s
ed
in
clin
ical
r
esear
ch
f
o
r
s
leep
s
tu
d
i
es,
s
p
ec
if
ically
f
o
r
d
etec
tin
g
s
leep
ap
n
ea
an
d
o
t
h
er
ca
r
d
io
v
ascu
lar
c
o
n
d
itio
n
s
[
2
0
]
.
T
h
e
h
ea
r
t r
ate
d
ata
in
th
is
d
ataset
in
clu
d
es:
−
I
n
s
tan
tan
eo
u
s
h
ea
r
t
r
ate
(
b
p
m
)
:
ex
tr
ac
ted
f
r
o
m
th
e
R
-
R
in
ter
v
al
s
in
E
C
G
s
ig
n
als
u
s
in
g
th
e
Pan
-
T
o
m
p
k
in
s
alg
o
r
ith
m
.
−
T
im
e
s
tam
p
:
th
e
tim
e
co
r
r
esp
o
n
d
in
g
to
ea
ch
h
ea
r
t
r
ate
m
e
asu
r
em
en
t,
with
r
ec
o
r
d
in
g
s
tak
en
at
in
ter
v
als
d
u
r
in
g
s
leep
.
T
h
e
d
ataset
s
p
an
s
s
ev
er
al
h
o
u
r
s
f
o
r
ea
c
h
in
d
i
v
id
u
al,
p
r
o
v
id
i
n
g
d
etailed
in
f
o
r
m
atio
n
o
n
h
e
ar
t
r
ate
v
ar
ia
b
ilit
y
th
r
o
u
g
h
o
u
t
th
e
s
leep
cy
cle.
T
h
e
p
r
im
ar
y
g
o
al
f
o
r
th
is
d
ataset
is
to
f
o
r
ec
ast
th
e
h
ea
r
t
r
ate
f
o
r
th
e
n
ex
t
2
m
in
u
tes,
b
ased
o
n
th
e
p
ast 3
0
m
in
u
tes o
f
o
b
s
er
v
e
d
h
ea
r
t
r
a
te
d
ata.
2
.
1
.
3
.
E
nerg
y
da
t
a
s
et
T
h
e
en
er
g
y
d
ataset
co
n
s
is
ts
o
f
h
o
u
r
ly
elec
tr
icity
co
n
s
u
m
p
tio
n
d
ata
f
r
o
m
th
e
Pen
n
s
y
l
v
an
ia
–
New
J
er
s
ey
–
Ma
r
y
lan
d
(
PJ
M
)
in
ter
c
o
n
n
ec
tio
n
p
o
wer
g
r
id
,
wh
ich
s
er
v
es
th
e
ea
s
ter
n
Un
ited
States
.
T
h
is
d
ataset
was
s
o
u
r
ce
d
f
r
o
m
th
e
p
u
b
licly
a
v
ailab
le
PJ
M
Ho
u
r
ly
L
o
ad
Da
ta
r
ep
o
s
ito
r
y
an
d
c
o
v
er
s
s
ev
er
al
y
ea
r
s
o
f
d
ata,
m
ak
in
g
it
well
-
s
u
ited
f
o
r
m
o
d
elin
g
lo
n
g
-
ter
m
tr
en
d
s
an
d
s
ea
s
o
n
al
p
atter
n
s
in
elec
tr
icity
d
e
m
an
d
.
T
h
e
d
ataset
in
clu
d
es th
e
f
o
llo
win
g
v
ar
ia
b
les:
−
T
o
tal
h
o
u
r
ly
lo
ad
(
MWh
)
:
th
e
to
tal
elec
tr
icity
d
em
an
d
i
n
m
e
g
awa
tt
-
h
o
u
r
s
f
o
r
ea
ch
h
o
u
r
.
−
Ho
u
r
o
f
d
ay
: th
e
s
p
ec
if
ic
h
o
u
r
o
f
th
e
d
a
y
th
e
m
ea
s
u
r
em
en
t w
as tak
en
.
−
Day
o
f
wee
k
:
a
ca
teg
o
r
ical
v
a
r
iab
le
in
d
icatin
g
t
h
e
d
a
y
o
f
th
e
wee
k
,
as
d
em
an
d
p
atter
n
s
v
ar
y
s
ig
n
if
ican
tly
b
etwe
en
wee
k
d
ay
s
an
d
wee
k
e
n
d
s
.
T
h
e
g
o
al
f
o
r
t
h
is
d
ataset
is
to
p
r
ed
ict
elec
tr
icity
d
em
a
n
d
f
o
r
th
e
n
ex
t
h
o
u
r
b
ased
o
n
th
e
p
r
ev
io
u
s
2
4
h
o
u
r
s
o
f
o
b
s
er
v
ed
d
ata
.
T
h
is
ty
p
e
o
f
f
o
r
ec
asti
n
g
is
cr
itical
f
o
r
g
r
id
o
p
er
ato
r
s
to
o
p
tim
ize
elec
tr
icity
g
en
er
atio
n
a
n
d
b
alan
ce
s
u
p
p
ly
with
d
em
an
d
,
esp
ec
ially
as r
en
ewa
b
le
en
er
g
y
s
o
u
r
ce
s
ar
e
in
te
g
r
ated
in
to
th
e
p
o
wer
g
r
id
.
2
.
2
.
P
re
pro
ce
s
s
ing
E
ac
h
d
ataset
u
n
d
er
we
n
t
s
p
ec
if
ic
p
r
ep
r
o
ce
s
s
in
g
s
tep
s
to
en
s
u
r
e
th
ey
wer
e
s
u
itab
le
f
o
r
tim
e
s
er
ies
f
o
r
ec
asti
n
g
.
−
Fin
an
ce
d
ata
p
r
ep
r
o
ce
s
s
in
g
:
s
to
ck
p
r
ices
wer
e
s
m
o
o
th
ed
u
s
in
g
a
Sav
itzk
y
-
Go
la
y
f
ilter
,
wh
i
ch
ap
p
lies
a
m
o
v
in
g
p
o
ly
n
o
m
ial
r
eg
r
es
s
io
n
to
th
e
d
ata.
T
h
e
f
ilter
u
s
ed
a
win
d
o
w
s
ize
o
f
5
d
ata
p
o
in
ts
an
d
a
p
o
ly
n
o
m
ial
o
r
d
e
r
o
f
2
,
wh
i
ch
en
s
u
r
e
d
m
i
n
im
al
d
is
to
r
tio
n
in
th
e
s
to
ck
p
r
ice
tr
en
d
s
.
T
h
e
d
ata
was
th
e
n
n
o
r
m
alize
d
u
s
in
g
m
in
-
m
a
x
n
o
r
m
aliza
tio
n
to
en
s
u
r
e
u
n
i
f
o
r
m
s
ca
lin
g
ac
r
o
s
s
all
v
ar
iab
les.
Miss
in
g
v
alu
es
wer
e
h
an
d
led
th
r
o
u
g
h
f
o
r
war
d
f
ill
tech
n
iq
u
es,
wh
er
e
th
e
m
o
s
t
r
ec
en
t
v
alid
o
b
s
er
v
ati
o
n
is
ca
r
r
ied
f
o
r
war
d
to
m
ain
tain
th
e
tem
p
o
r
al
s
eq
u
en
ce
an
d
e
n
s
u
r
e
n
o
g
a
p
s
in
th
e
tim
e
s
er
ies.
−
Hea
lth
ca
r
e
d
ata
p
r
ep
r
o
ce
s
s
in
g
:
th
e
Pan
-
T
o
m
p
k
in
s
al
g
o
r
it
h
m
was
ap
p
lied
t
o
th
e
r
aw
E
C
G
s
ig
n
als
to
ex
tr
ac
t
in
s
tan
tan
eo
u
s
h
ea
r
t
r
at
e
b
y
d
etec
tin
g
R
-
p
ea
k
s
in
th
e
QR
S
co
m
p
lex
[
21]
.
On
ce
th
e
R
-
p
ea
k
s
wer
e
id
en
tifie
d
,
t
h
e
R
-
R
in
ter
v
als
wer
e
u
s
ed
t
o
ca
lcu
late
th
e
i
n
s
tan
tan
eo
u
s
h
ea
r
t
r
ate
(
b
p
m
)
.
T
o
en
s
u
r
e
ev
e
n
ly
s
p
ac
ed
tim
e
in
te
r
v
als,
th
e
h
ea
r
t
r
ate
d
ata
was
in
ter
p
o
lat
ed
u
s
in
g
cu
b
ic
s
p
lin
e
i
n
ter
p
o
latio
n
,
wh
ic
h
m
ain
tain
s
s
m
o
o
th
tr
an
s
itio
n
s
b
etwe
en
d
ata
p
o
in
ts
.
Ou
tlier
s
—
d
ef
in
ed
as
h
ea
r
t r
ate
v
alu
es b
elo
w
3
0
b
p
m
o
r
ab
o
v
e
2
0
0
b
p
m
—
wer
e
id
e
n
tifie
d
an
d
r
em
o
v
ed
to
av
o
id
p
h
y
s
io
lo
g
ical
an
o
m
alies.
Fin
ally
,
th
e
h
e
ar
t
r
ate
d
ata
was n
o
r
m
alize
d
u
s
in
g
m
i
n
-
m
ax
n
o
r
m
aliza
tio
n
.
−
E
n
er
g
y
d
ata
p
r
ep
r
o
ce
s
s
in
g
:
th
e
en
er
g
y
d
ataset
was
d
etr
en
d
e
d
u
s
in
g
a
Sav
itzk
y
-
Go
la
y
f
ilte
r
with
a
win
d
o
w
s
ize
o
f
2
4
h
o
u
r
s
an
d
a
p
o
ly
n
o
m
ial
o
r
d
er
o
f
2
.
T
h
is
f
ilter
h
elp
ed
s
m
o
o
th
o
u
t
d
aily
s
ea
s
o
n
ality
p
atter
n
s
wh
ile
p
r
eser
v
in
g
im
p
o
r
tan
t
co
n
s
u
m
p
tio
n
tr
en
d
s
.
T
h
e
d
ata
was
th
en
n
o
r
m
alize
d
u
s
in
g
m
in
-
m
ax
n
o
r
m
aliza
tio
n
to
en
s
u
r
e
u
n
if
o
r
m
s
ca
lin
g
o
f
elec
tr
icity
c
o
n
s
u
m
p
tio
n
v
alu
es.
Miss
in
g
v
alu
es
wer
e
im
p
u
ted
u
s
in
g
lin
ea
r
in
ter
p
o
latio
n
,
wh
er
e
m
is
s
in
g
d
ata
p
o
in
ts
wer
e
esti
m
ated
b
ased
o
n
th
e
v
alu
es
im
m
ed
iately
b
ef
o
r
e
an
d
af
ter
th
e
g
ap
.
A
r
o
llin
g
win
d
o
w
tech
n
iq
u
e
was
ap
p
lied
to
ca
p
tu
r
e
s
ea
s
o
n
ality
an
d
v
ar
iatio
n
s
in
d
em
an
d
p
atter
n
s
o
v
er
tim
e,
allo
win
g
th
e
m
o
d
els
to
ad
ju
s
t
to
r
ec
u
r
r
in
g
cy
cles
in
th
e
d
ata.
E
ac
h
d
ataset
was
s
p
lit
in
to
tr
ain
in
g
(
7
0
%),
v
alid
atio
n
(
1
5
%),
an
d
test
s
ets
(
1
5
%).
T
h
e
v
alid
atio
n
s
et
was
u
s
ed
f
o
r
h
y
p
er
p
ar
am
eter
tu
n
in
g
,
wh
ile
th
e
test
s
et
was r
eser
v
ed
f
o
r
ev
alu
atin
g
m
o
d
el
p
er
f
o
r
m
an
ce
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
7
2
2
-
3
2
2
1
C
o
m
p
u
t Sci
I
n
f
T
ec
h
n
o
l
,
Vo
l.
7
,
No
.
3
,
No
v
em
b
er
20
26
:
2
9
1
-
303
294
2
.
3
.
M
o
del
im
plem
ent
a
t
io
n
T
h
r
ee
f
o
r
ec
asti
n
g
m
o
d
els
wer
e
im
p
lem
en
ted
:
AR
I
MA
,
L
STM
,
an
d
T
FT.
E
ac
h
m
o
d
el
was f
in
e
-
tu
n
e
d
f
o
r
o
p
tim
al
p
e
r
f
o
r
m
an
ce
b
ased
o
n
th
e
s
p
ec
if
ic
ch
ar
ac
te
r
is
tics
o
f
th
e
d
atasets
(
f
in
an
c
e,
h
ea
lth
ca
r
e,
an
d
en
er
g
y
)
.
T
h
e
h
y
p
e
r
p
ar
am
ete
r
s
f
o
r
all
m
o
d
els we
r
e
s
elec
ted
b
ased
o
n
g
r
i
d
s
ea
r
ch
u
s
in
g
th
e
v
alid
atio
n
s
et.
2
.
3
.
1
.
Aut
o
re
g
re
s
s
iv
e
inte
g
ra
t
ed
m
o
v
ing
a
v
er
a
g
e
mo
del
T
h
e
AR
I
MA
m
o
d
el,
a
tr
ad
itio
n
al
s
tatis
tica
l
m
eth
o
d
f
o
r
tim
e
s
er
ies
f
o
r
ec
asti
n
g
,
h
as
b
ee
n
wi
d
ely
u
s
ed
in
v
ar
i
o
u
s
d
o
m
ain
s
d
u
e
to
its
s
im
p
licity
an
d
ef
f
icien
c
y
in
h
an
d
lin
g
lin
ea
r
r
elatio
n
s
h
ip
s
.
T
h
e
AR
I
MA
m
o
d
el
co
m
p
r
is
es
th
r
ee
c
o
m
p
o
n
en
ts
:
au
to
r
eg
r
ess
io
n
(
AR
)
,
d
if
f
er
e
n
cin
g
(
I
)
,
a
n
d
m
o
v
in
g
av
e
r
a
g
e
(
MA
)
.
T
h
e
AR
co
m
p
o
n
en
t c
ap
tu
r
es th
e
lin
ea
r
r
elatio
n
s
h
ip
b
etwe
en
p
ast an
d
cu
r
r
en
t v
alu
es,
wh
ile
th
e
MA
co
m
p
o
n
en
t m
o
d
els
th
e
d
ep
en
d
en
cy
b
etwe
en
th
e
c
u
r
r
en
t v
al
u
e
an
d
p
ast f
o
r
ec
ast
er
r
o
r
s
.
T
h
e
I
co
m
p
o
n
e
n
t in
v
o
l
v
es d
if
f
er
en
cin
g
th
e
ti
m
e
s
er
ies to
ac
h
iev
e
s
tatio
n
ar
ity
.
I
n
th
is
s
tu
d
y
,
th
e
AR
I
MA
m
o
d
el
was
co
n
f
ig
u
r
e
d
b
y
s
elec
tin
g
th
e
o
p
tim
al
p
ar
am
eter
s
(
p
,
d
,
an
d
q
)
u
s
in
g
th
e
B
o
x
-
J
en
k
in
s
m
eth
o
d
o
lo
g
y
.
T
h
e
B
o
x
-
J
en
k
in
s
m
eth
o
d
o
l
o
g
y
is
a
s
y
s
tem
atic
ap
p
r
o
ac
h
u
s
ed
to
id
en
tify
,
esti
m
ate,
an
d
d
iag
n
o
s
e
A
R
I
MA
m
o
d
els.
T
h
e
p
r
o
ce
s
s
in
v
o
lv
es
an
aly
zin
g
au
to
co
r
r
elatio
n
an
d
p
ar
tial
au
to
co
r
r
elatio
n
f
u
n
ctio
n
s
(
AC
F
an
d
PA
C
F)
to
d
eter
m
in
e
th
e
ap
p
r
o
p
r
iate
o
r
d
er
o
f
th
e
AR
an
d
MA
co
m
p
o
n
en
ts
,
co
n
d
u
ctin
g
s
tatio
n
ar
ity
test
s
to
d
eter
m
in
e
th
e
d
if
f
er
en
cin
g
(
I
)
o
r
d
er
,
an
d
ev
alu
atin
g
m
o
d
el
d
iag
n
o
s
tics
to
e
n
s
u
r
e
t
h
e
m
o
d
el's
ad
eq
u
ac
y
.
AC
F
an
d
PAC
F
p
lo
ts
wer
e
an
aly
ze
d
to
d
eter
m
in
e
th
e
ap
p
r
o
p
r
iate
v
alu
es
f
o
r
p
an
d
q
,
wh
ile
th
e
d
if
f
e
r
en
cin
g
(
I
)
p
ar
am
et
er
d
was
s
elec
ted
b
ased
o
n
s
tatio
n
ar
ity
test
s
.
T
h
e
h
y
p
er
p
ar
a
m
eter
s
wer
e
o
p
tim
ized
b
y
m
in
im
izin
g
th
e
A
k
aik
e
in
f
o
r
m
atio
n
c
r
iter
io
n
(
AI
C
)
,
an
d
th
e
f
i
n
al
v
alu
es we
r
e
s
elec
ted
b
ased
o
n
th
e
v
alid
atio
n
s
et
,
as d
etailed
i
n
T
ab
le
1
.
2
.
3
.
2
.
L
o
ng
s
ho
rt
-
t
er
m m
e
mo
ry
net
wo
r
k
L
STM
n
etwo
r
k
s
,
a
ty
p
e
o
f
R
N
Ns,
h
av
e
em
er
g
ed
as a
p
o
wer
f
u
l to
o
l f
o
r
tim
e
s
er
ies f
o
r
ec
asti
n
g
d
u
e
to
th
eir
ab
ilit
y
to
ca
p
t
u
r
e
lo
n
g
-
t
er
m
d
ep
e
n
d
en
cies
an
d
n
o
n
-
li
n
ea
r
p
atter
n
s
in
s
eq
u
en
tial
d
a
ta.
L
STM
s
p
ar
tially
ad
d
r
ess
th
e
v
an
is
h
in
g
g
r
a
d
ien
t
p
r
o
b
lem
,
a
co
m
m
o
n
i
s
s
u
e
in
t
r
ad
itio
n
al
R
NNs,
b
y
in
co
r
p
o
r
a
tin
g
m
em
o
r
y
ce
lls
an
d
g
atin
g
m
ec
h
an
is
m
s
th
at
co
n
tr
o
l
th
e
f
lo
w
o
f
i
n
f
o
r
m
atio
n
th
r
o
u
g
h
th
e
n
etwo
r
k
.
T
h
e
L
STM
ar
ch
itectu
r
e
ty
p
ically
co
n
s
is
ts
o
f
m
u
ltip
le
L
STM
lay
er
s
,
ea
ch
co
n
tain
in
g
a
s
et
o
f
L
STM
u
n
its
.
E
ac
h
L
STM
u
n
it
co
m
p
r
is
es
a
m
em
o
r
y
ce
ll,
an
in
p
u
t g
ate,
an
o
u
tp
u
t g
ate,
an
d
a
f
o
r
g
et
g
a
te.
T
h
e
m
em
o
r
y
ce
ll st
o
r
es in
f
o
r
m
atio
n
o
v
er
tim
e
,
wh
ile
th
e
g
ates
r
eg
u
late
th
e
f
lo
w
o
f
in
f
o
r
m
atio
n
in
to
a
n
d
o
u
t
o
f
th
e
ce
ll.
T
h
e
in
p
u
t
g
ate
d
e
ter
m
in
es
h
o
w
m
u
c
h
n
ew
in
f
o
r
m
atio
n
is
s
to
r
ed
i
n
th
e
ce
ll,
th
e
o
u
t
p
u
t
g
ate
co
n
tr
o
ls
h
o
w
m
u
ch
in
f
o
r
m
atio
n
is
p
ass
ed
to
th
e
n
ex
t
lay
er
,
an
d
th
e
f
o
r
g
et
g
ate
d
ec
i
d
es wh
at
in
f
o
r
m
atio
n
is
d
is
ca
r
d
ed
f
r
o
m
th
e
ce
ll.
I
n
th
is
s
tu
d
y
,
th
e
L
STM
ar
c
h
itectu
r
e
co
n
s
is
ted
o
f
two
L
STM
lay
er
s
f
o
llo
wed
b
y
a
d
en
s
e
o
u
tp
u
t
lay
er
.
Dr
o
p
o
u
t
r
e
g
u
lar
izatio
n
was
ap
p
lied
to
p
r
e
v
en
t
o
v
e
r
f
itti
n
g
,
a
n
d
th
e
A
d
am
o
p
ti
m
izer
was
u
s
ed
to
m
in
im
ize
th
e
lo
s
s
f
u
n
ctio
n
.
H
y
p
er
p
a
r
am
eter
s
s
u
ch
as
th
e
n
u
m
b
er
o
f
L
STM
u
n
its
,
lea
r
n
in
g
r
ate,
d
r
o
p
o
u
t
r
ate
,
an
d
b
atch
s
ize
wer
e
f
in
e
-
tu
n
e
d
u
s
in
g
g
r
id
s
ea
r
ch
,
w
ith
th
e
f
in
al
v
alu
es
s
elec
ted
b
ased
o
n
th
e
v
alid
atio
n
s
et
,
r
ef
er
to
T
a
b
le
1
f
o
r
d
etails.
2
.
3
.
3
.
T
em
po
ra
l f
us
io
n t
ra
ns
f
o
rm
er
W
e
im
p
lem
en
ted
th
e
T
FT
f
o
r
tim
e
s
er
ies
f
o
r
ec
asti
n
g
ac
r
o
s
s
f
in
an
ce
,
h
ea
lth
ca
r
e,
an
d
e
n
er
g
y
d
atasets
.
T
h
e
m
o
d
el
f
o
llo
ws
an
e
n
co
d
er
-
d
ec
o
d
er
s
tr
u
ctu
r
e
,
in
co
r
p
o
r
atin
g
s
p
ec
ialized
co
m
p
o
n
e
n
t
s
s
u
ch
as
v
ar
iab
le
s
elec
tio
n
n
etwo
r
k
s
(
VSNs
)
,
g
ated
r
esid
u
al
n
etwo
r
k
s
(
GR
Ns),
an
d
tem
p
o
r
al
s
elf
-
atten
t
io
n
[
2
2
]
.
T
h
e
VSN
d
y
n
am
ically
s
elec
ts
r
elev
an
t
f
ea
tu
r
es
at
ea
c
h
tim
e
s
tep
.
I
t
p
r
o
ce
s
s
es
in
p
u
ts
u
s
in
g
f
u
ll
y
co
n
n
ec
ted
la
y
er
s
,
ap
p
lies
a
s
p
ar
s
em
ax
ac
tiv
atio
n
f
u
n
ctio
n
to
en
f
o
r
ce
s
p
ar
s
ity
,
an
d
ass
ig
n
s
lear
n
ed
im
p
o
r
tan
ce
weig
h
ts
to
ea
ch
v
ar
iab
le.
T
h
is
en
s
u
r
es th
at
o
n
l
y
th
e
m
o
s
t r
elev
a
n
t in
p
u
ts
co
n
tr
ib
u
te
to
f
o
r
ec
asti
n
g
.
T
h
e
en
co
d
e
r
p
r
o
ce
s
s
es
h
is
to
r
ical
in
p
u
ts
u
s
i
n
g
L
STM
n
etwo
r
k
s
,
wh
ich
ca
p
tu
r
e
tem
p
o
r
al
d
ep
en
d
e
n
cies.
T
h
e
GR
Ns
r
ef
in
e
th
ese
en
co
d
ed
r
ep
r
esen
tatio
n
s
b
ef
o
r
e
p
ass
in
g
th
em
to
s
u
b
s
eq
u
en
t
lay
er
s
.
E
ac
h
GR
N
co
n
s
is
ts
o
f
f
u
lly
co
n
n
ec
ted
la
y
er
s
,
ex
p
o
n
e
n
tial
lin
ea
r
u
n
it
(
ELU
)
ac
tiv
atio
n
f
u
n
ctio
n
s
,
g
atin
g
m
ec
h
an
is
m
s
(
s
ig
m
o
id
ac
tiv
ati
o
n
s
)
,
an
d
r
esid
u
al
c
o
n
n
ec
tio
n
s
to
r
eg
u
late
in
f
o
r
m
atio
n
f
lo
w.
GR
Ns
ar
e
also
u
s
ed
in
th
e
d
ec
o
d
er
to
p
r
o
ce
s
s
k
n
o
wn
f
u
tu
r
e
c
o
v
ar
iates.
T
o
ca
p
tu
r
e
lo
n
g
-
r
an
g
e
d
ep
en
d
en
cies,
th
e
m
o
d
el
ap
p
lies
m
ask
ed
in
ter
p
r
etab
le
m
u
lti
-
h
ea
d
s
elf
-
atten
tio
n
to
t
h
e
en
co
d
ed
r
ep
r
esen
tatio
n
s
.
T
h
is
m
ec
h
an
i
s
m
co
m
p
u
tes
q
u
er
y
(
Q)
,
k
e
y
(
K)
,
an
d
v
al
u
e
(
V)
m
atr
ices
f
r
o
m
th
e
L
STM
o
u
tp
u
ts
an
d
lear
n
s
tem
p
o
r
al
r
elati
o
n
s
h
ip
s
with
o
u
t
lo
o
k
in
g
i
n
to
t
h
e
f
u
tu
r
e
(
en
f
o
r
ce
d
v
ia
ca
u
s
al
m
ask
in
g
)
.
T
h
e
n
u
m
b
er
o
f
atten
tio
n
h
ea
d
s
an
d
d
r
o
p
o
u
t
r
ates
f
o
r
th
is
m
ec
h
an
is
m
ar
e
lis
ted
in
T
ab
le
1
.
Af
ter
s
elf
-
atten
tio
n
,
ad
d
itio
n
al
GR
N
s
r
ef
in
e
th
e
lear
n
ed
r
ep
r
esen
tatio
n
s
b
ef
o
r
e
g
en
er
atin
g
f
in
al
f
o
r
ec
asts
.
T
h
e
o
u
tp
u
t
lay
e
r
p
r
o
d
u
ce
s
q
u
a
n
tile
esti
m
ates
(
1
0
th
,
5
0
t
h
,
a
n
d
9
0
th
p
e
r
ce
n
tiles
)
to
q
u
an
tify
u
n
ce
r
ta
in
ty
.
Sin
ce
T
FT
o
u
tp
u
ts
q
u
an
tile
f
o
r
ec
asts
,
th
e
f
in
al
p
o
in
t
esti
m
ate
was
tak
en
as
th
e
5
0
t
h
p
er
ce
n
tile
(
m
ed
ian
p
r
e
d
ictio
n
)
,
wh
ic
h
r
e
p
r
esen
ts
th
e
m
o
s
t
lik
ely
o
u
tco
m
e.
Fo
r
d
ec
is
io
n
-
m
a
k
in
g
a
p
p
licatio
n
s
r
eq
u
ir
in
g
Evaluation Warning : The document was created with Spire.PDF for Python.
C
o
m
p
u
t Sci
I
n
f
T
ec
h
n
o
l
I
SS
N:
2722
-
3
2
2
1
Time
s
er
ies fo
r
ec
a
s
tin
g
:
a
co
mp
a
r
a
tive
a
n
a
lysi
s
o
f A
R
I
MA,
LS
TM
,
a
n
d
TFT m
o
d
els
…
(
Ma
r
ya
m
Ho
s
s
ein
i
)
295
r
is
k
ass
ess
m
en
t,
th
e
1
0
th
an
d
9
0
th
p
er
ce
n
tiles
c
an
b
e
u
s
ed
to
p
r
o
v
id
e
lo
wer
an
d
u
p
p
er
co
n
f
id
en
ce
b
o
u
n
d
s
.
T
h
e
m
o
d
el
was
tr
ain
ed
u
s
in
g
th
e
Ad
am
o
p
tim
izer
,
with
a
q
u
an
tile
lo
s
s
f
u
n
ctio
n
to
o
p
tim
i
ze
f
o
r
ec
asts
ac
r
o
s
s
d
if
f
er
en
t
q
u
a
n
tile
lev
els.
Hy
p
er
p
ar
am
eter
s
s
u
ch
as
th
e
le
ar
n
in
g
r
ate,
b
atch
s
ize,
n
u
m
b
er
o
f
GR
N
lay
er
s
,
h
id
d
en
u
n
its
,
an
d
d
r
o
p
o
u
t
r
ate
wer
e
o
p
tim
ized
v
ia
g
r
id
s
ea
r
c
h
as sh
o
wn
in
T
a
b
le
1
.
Fo
r
ea
ch
d
ataset,
f
ea
tu
r
e
en
g
in
ee
r
in
g
was
p
er
f
o
r
m
e
d
to
p
r
e
p
ar
e
th
e
d
ata
f
o
r
th
e
T
FT
m
o
d
el.
I
n
th
e
f
in
an
ce
d
ataset,
r
aw
s
to
ck
p
r
i
ce
s
(
o
p
en
,
clo
s
e,
h
ig
h
,
a
n
d
lo
w)
an
d
v
o
lu
m
e
wer
e
d
ir
ec
tly
u
s
ed
as
o
b
s
er
v
ed
h
is
to
r
ical
in
p
u
ts
,
with
n
o
s
tatic
o
r
k
n
o
wn
f
u
tu
r
e
in
p
u
ts
i
n
th
is
b
asic
s
etu
p
.
Fo
r
th
e
h
ea
lth
ca
r
e
d
ataset,
th
e
in
s
tan
tan
eo
u
s
h
ea
r
t
r
ate
w
as
th
e
p
r
im
ar
y
o
b
s
er
v
ed
h
is
to
r
ical
in
p
u
t,
wh
ile
t
h
e
tim
e
s
tam
p
was
tr
an
s
f
o
r
m
ed
in
to
m
in
u
te
-
of
-
d
ay
f
ea
tu
r
es
f
o
r
b
o
th
o
b
s
er
v
ed
h
is
to
r
ical
an
d
k
n
o
wn
f
u
tu
r
e
in
p
u
ts
.
T
h
e
en
e
r
g
y
d
ataset
lev
er
ag
ed
t
o
tal
h
o
u
r
ly
lo
a
d
a
s
th
e
co
r
e
o
b
s
er
v
ed
h
is
to
r
ical
in
p
u
t,
alo
n
g
s
id
e
h
o
u
r
-
of
-
d
a
y
an
d
d
ay
-
of
-
wee
k
f
ea
tu
r
es
f
o
r
b
o
t
h
h
is
to
r
ical
a
n
d
f
u
tu
r
e
c
o
n
tex
ts
.
I
n
all
d
atasets
,
tim
e
s
tam
p
s
wer
e
co
n
v
er
ted
in
to
r
elev
a
n
t
tem
p
o
r
al
f
ea
tu
r
es
to
ca
p
tu
r
e
s
ea
s
o
n
ality
an
d
cy
clica
l
p
atter
n
s
.
Fu
r
th
er
m
o
r
e,
ca
teg
o
r
ical
v
a
r
iab
les,
lik
e
d
ay
-
of
-
wee
k
,
wer
e
o
n
e
-
h
o
t e
n
co
d
e
d
t
o
f
ac
ilit
ate
m
o
d
el
p
r
o
ce
s
s
in
g
.
2
.
4
.
M
o
del
t
ra
ini
n
g
a
nd
v
a
li
da
t
io
n
All
f
o
r
ec
asti
n
g
m
o
d
els
wer
e
tr
ain
ed
u
s
in
g
th
e
t
r
ain
in
g
s
u
b
s
et,
wh
ile
h
y
p
er
p
ar
a
m
eter
o
p
tim
izatio
n
was
p
er
f
o
r
m
ed
o
n
th
e
v
alid
at
io
n
s
u
b
s
et.
T
o
en
s
u
r
e
r
ea
lis
tic
tim
e
-
s
er
ies
f
o
r
ec
asti
n
g
an
d
p
r
ev
en
t
in
f
o
r
m
atio
n
leak
ag
e,
r
o
llin
g
win
d
o
w
cr
o
s
s
-
v
ali
d
atio
n
was
em
p
lo
y
e
d
ac
r
o
s
s
all
d
ataset
s
,
in
clu
d
in
g
f
in
a
n
ce
,
h
ea
lth
ca
r
e,
a
n
d
en
er
g
y
[
2
3
]
.
Un
lik
e
c
o
n
v
e
n
tio
n
al
r
an
d
o
m
s
am
p
lin
g
s
tr
ateg
ies,
r
o
llin
g
win
d
o
w
v
alid
atio
n
p
r
eser
v
es
tem
p
o
r
al
o
r
d
er
in
g
b
y
en
s
u
r
in
g
th
at
f
u
t
u
r
e
o
b
s
er
v
atio
n
s
ar
e
n
ev
e
r
u
s
ed
d
u
r
in
g
m
o
d
el
tr
ain
in
g
.
T
h
is
p
r
o
ce
d
u
r
e
m
o
r
e
clo
s
ely
r
ef
lects
r
ea
l
-
wo
r
ld
f
o
r
ec
asti
n
g
co
n
d
itio
n
s
,
wh
er
e
p
r
ed
ictio
n
s
m
u
s
t
r
ely
s
o
lely
o
n
h
is
to
r
ical
in
f
o
r
m
atio
n
.
Fo
r
ea
ch
d
ataset,
th
e
o
b
s
er
v
at
io
n
s
wer
e
f
ir
s
t
ch
r
o
n
o
lo
g
ically
d
iv
id
e
d
in
to
tr
ai
n
in
g
(
7
0
%),
v
alid
atio
n
(
1
5
%),
an
d
test
in
g
(
1
5
%)
s
u
b
s
ets.
T
h
e
tr
ain
in
g
s
et
was
u
s
ed
to
f
it
th
e
m
o
d
els,
wh
il
e
th
e
v
alid
atio
n
s
et
s
u
p
p
o
r
ted
h
y
p
er
p
ar
am
eter
o
p
tim
izatio
n
an
d
m
o
d
el
s
elec
tio
n
.
T
h
e
test
in
g
s
et
r
em
ain
ed
co
m
p
letely
u
n
s
ee
n
d
u
r
in
g
m
o
d
el
d
e
v
elo
p
m
e
n
t
a
n
d
was
r
eser
v
e
d
ex
cl
u
s
iv
ely
f
o
r
f
in
al
p
er
f
o
r
m
an
ce
ass
es
s
m
en
t.
Ma
in
tain
in
g
ch
r
o
n
o
lo
g
ical
s
ep
ar
atio
n
en
s
u
r
ed
p
r
eser
v
atio
n
o
f
tem
p
o
r
al
d
ep
en
d
e
n
cies
an
d
p
r
ev
e
n
ted
f
u
tu
r
e
o
b
s
er
v
atio
n
s
f
r
o
m
in
f
lu
en
cin
g
m
o
d
el
co
n
s
tr
u
ctio
n
.
T
o
f
u
r
th
er
im
p
r
o
v
e
r
o
b
u
s
tn
e
s
s
an
d
r
ed
u
ce
d
ep
en
d
e
n
ce
o
n
a
s
in
g
le
tem
p
o
r
al
s
p
lit,
an
ex
p
an
d
in
g
r
o
llin
g
win
d
o
w
c
r
o
s
s
-
v
alid
at
io
n
s
tr
ateg
y
was
ap
p
lied
d
u
r
in
g
tr
ai
n
in
g
an
d
v
alid
atio
n
.
I
n
th
is
ap
p
r
o
ac
h
,
th
e
tr
ain
in
g
win
d
o
w
p
r
o
g
r
ess
iv
ely
in
cr
ea
s
ed
in
s
ize
at
ea
ch
iter
atio
n
wh
ile
v
alid
atio
n
was
co
n
d
u
cted
o
n
th
e
im
m
ed
iately
f
o
llo
win
g
u
n
s
ee
n
o
b
s
er
v
atio
n
s
.
Fo
r
ex
am
p
le,
f
o
r
th
e
en
er
g
y
d
ataset,
th
e
f
ir
s
t
iter
atio
n
u
s
ed
s
am
p
les
T
1
–
T
1
0
0
0
f
o
r
tr
ain
i
n
g
an
d
T
1
0
0
1
–
T
1
1
0
0
f
o
r
v
alid
atio
n
.
I
n
th
e
n
e
x
t
iter
atio
n
,
th
e
tr
ain
in
g
win
d
o
w
ex
p
an
d
e
d
to
T
1
–
T
1
1
0
0
wh
ile
th
e
v
alid
atio
n
p
er
io
d
s
h
if
ted
to
T
1
1
0
1
–
T
1
2
0
0
.
T
h
is
p
r
o
ce
d
u
r
e
was
r
e
p
ea
ted
ac
r
o
s
s
s
u
cc
ess
iv
e
win
d
o
ws u
n
til th
e
en
d
o
f
th
e
tr
ain
in
g
-
v
alid
atio
n
p
er
io
d
was r
ea
ch
ed
.
T
h
e
s
am
e
s
tr
ateg
y
wa
s
ap
p
lied
to
th
e
f
in
a
n
cial
an
d
h
ea
lth
ca
r
e
d
atasets
.
T
h
e
ex
p
an
d
in
g
-
win
d
o
w
d
esig
n
en
ab
led
r
ep
ea
ted
ev
al
u
atio
n
o
n
f
u
tu
r
e
u
n
s
ee
n
o
b
s
er
v
atio
n
s
an
d
p
r
o
v
i
d
ed
a
m
o
r
e
r
eliab
le
esti
m
ate
o
f
m
o
d
el
g
en
er
ali
za
tio
n
p
er
f
o
r
m
an
ce
u
n
d
er
p
r
ac
tical
f
o
r
ec
asti
n
g
co
n
d
itio
n
s
.
Hy
p
er
p
ar
a
m
eter
o
p
tim
izatio
n
was
co
n
d
u
cted
u
s
in
g
g
r
id
s
ea
r
ch
ac
r
o
s
s
p
r
ed
e
f
in
ed
p
ar
am
e
ter
r
an
g
es
s
h
o
wn
in
T
ab
le
1
.
Fo
r
AR
I
MA
,
co
m
b
in
atio
n
s
o
f
au
to
r
eg
r
es
s
iv
e
(
p
)
,
d
if
f
e
r
en
cin
g
(
d
)
,
a
n
d
m
o
v
in
g
av
er
a
g
e
(
q
)
p
ar
am
eter
s
wer
e
s
y
s
tem
atica
lly
ex
p
lo
r
ed
an
d
ev
alu
ate
d
u
s
in
g
th
e
AI
C
.
Fo
r
L
S
T
M
an
d
T
FT,
h
y
p
e
r
p
ar
am
ete
r
s
in
clu
d
in
g
lear
n
in
g
r
ate,
h
id
d
e
n
lay
er
s
ize,
b
atch
s
ize,
d
r
o
p
o
u
t
r
ate,
an
d
n
etwo
r
k
a
r
ch
itect
u
r
e
p
ar
am
ete
r
s
wer
e
ev
alu
ated
ac
r
o
s
s
ca
n
d
id
ate
v
a
lu
es.
T
h
e
o
p
tim
al
co
n
f
ig
u
r
atio
n
was
s
elec
ted
b
ased
o
n
v
alid
atio
n
p
er
f
o
r
m
an
ce
,
p
r
im
ar
ily
m
in
im
izin
g
MA
E
an
d
MA
PE
wh
ile
m
ax
im
izin
g
R
²
.
T
o
im
p
r
o
v
e
s
tatis
tical
r
o
b
u
s
tn
ess
an
d
r
e
d
u
ce
v
ar
iab
ilit
y
ar
i
s
in
g
f
r
o
m
s
to
ch
asti
c
tr
ai
n
in
g
p
r
o
ce
s
s
es,
ea
ch
d
ee
p
lear
n
i
n
g
e
x
p
er
im
en
t
(
L
STM
an
d
T
FT)
was
r
ep
ea
ted
m
u
ltip
le
tim
es
u
s
in
g
d
if
f
er
en
t
r
an
d
o
m
in
itializatio
n
s
ee
d
s
,
an
d
av
er
ag
e
p
er
f
o
r
m
an
ce
m
etr
ics
wer
e
r
ep
o
r
ted
.
T
h
is
ap
p
r
o
a
ch
r
e
d
u
ce
d
s
en
s
itiv
ity
to
r
an
d
o
m
weig
h
t
in
itializatio
n
an
d
tr
ain
in
g
f
l
u
ctu
atio
n
s
.
F
u
r
th
er
m
o
r
e,
ea
r
ly
s
to
p
p
in
g
was
im
p
lem
en
ted
t
o
p
r
ev
en
t
o
v
er
f
itti
n
g
,
ter
m
in
ati
n
g
tr
ain
in
g
if
v
alid
atio
n
l
o
s
s
f
ailed
to
im
p
r
o
v
e
o
v
er
ten
co
n
s
ec
u
tiv
e
ep
o
c
h
s
.
Af
ter
o
p
tim
al
h
y
p
er
p
a
r
am
eter
s
h
ad
b
ee
n
id
en
tifie
d
,
m
o
d
els
wer
e
r
etr
ain
ed
u
s
in
g
th
e
co
m
b
in
ed
tr
ain
in
g
an
d
v
alid
atio
n
d
atasets
to
m
ax
im
i
ze
d
ata
u
tili
za
tio
n
.
Fin
ally
,
m
o
d
el
p
er
f
o
r
m
a
n
ce
was
ev
alu
at
ed
o
n
th
e
h
eld
-
o
u
t
test
d
ataset,
wh
ich
co
n
s
titu
ted
th
e
f
in
al
1
5
%
o
f
o
b
s
er
v
atio
n
s
an
d
was
n
o
t
i
n
v
o
lv
ed
in
eit
h
e
r
m
o
d
el
tr
ain
in
g
o
r
h
y
p
er
p
ar
am
eter
s
elec
tio
n
.
T
h
i
s
s
tr
ict
s
ep
ar
atio
n
en
s
u
r
ed
u
n
b
iased
ass
ess
m
en
t
o
f
p
r
ed
ictiv
e
p
er
f
o
r
m
an
ce
an
d
en
h
an
ce
d
th
e
r
ep
r
o
d
u
cib
ilit
y
a
n
d
r
e
liab
ilit
y
o
f
th
e
r
e
p
o
r
ted
r
esu
lts
.
2
.
5
.
Sens
it
iv
it
y
a
na
l
y
s
is
o
f
mis
s
i
ng
da
t
a
T
o
ass
es
s
th
e
s
en
s
itiv
ity
o
f
o
u
r
m
o
d
els
to
m
is
s
in
g
d
ata,
w
e
co
n
d
u
cted
a
co
m
p
r
eh
en
s
iv
e
an
aly
s
is
ac
r
o
s
s
all
th
r
ee
d
ataset
s
:
f
in
an
cial,
h
ea
lth
ca
r
e,
an
d
en
er
g
y
.
W
e
r
ec
o
g
n
ized
th
at
th
e
im
p
ac
t
o
f
m
is
s
in
g
d
ata
ca
n
v
ar
y
s
ig
n
if
ican
tly
d
ep
en
d
in
g
o
n
th
e
s
p
ec
if
ic
ch
ar
ac
ter
is
tics
o
f
ea
ch
d
ataset
an
d
th
e
m
ec
h
a
n
i
s
m
s
b
y
wh
ich
d
ata
b
ec
o
m
es m
is
s
in
g
.
T
h
er
e
f
o
r
e,
we
tailo
r
ed
o
u
r
ap
p
r
o
ac
h
t
o
r
e
f
lect
th
ese
n
u
an
ce
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
7
2
2
-
3
2
2
1
C
o
m
p
u
t Sci
I
n
f
T
ec
h
n
o
l
,
Vo
l.
7
,
No
.
3
,
No
v
em
b
er
20
26
:
2
9
1
-
303
296
I
n
th
e
f
in
a
n
cial
d
at
aset,
wh
er
e
m
is
s
in
g
d
ata
m
ig
h
t
ar
is
e
f
r
o
m
in
f
r
eq
u
en
t
ev
e
n
ts
lik
e
tr
ad
i
n
g
h
alts
o
r
d
ata
er
r
o
r
s
,
we
i
n
tr
o
d
u
ce
d
lo
w
er
lev
els
o
f
m
is
s
in
g
n
ess
,
r
a
n
g
i
n
g
f
r
o
m
1
%
to
5
%.
C
o
n
v
er
s
el
y
,
in
th
e
h
ea
lth
ca
r
e
d
ataset,
wh
er
e
m
is
s
in
g
d
ata
c
o
u
ld
b
e
m
o
r
e
p
r
ev
alen
t
d
u
e
t
o
f
ac
to
r
s
s
u
ch
as
s
en
s
o
r
m
alf
u
n
ctio
n
s
o
r
p
atien
t
m
o
v
em
en
ts
,
we
ex
p
lo
r
ed
a
b
r
o
ad
er
r
an
g
e
o
f
m
is
s
in
g
n
ess
,
f
r
o
m
5
%
to
2
0
%.
Fo
r
t
h
e
en
e
r
g
y
d
ataset,
wh
er
e
m
is
s
in
g
d
ata
m
ig
h
t
s
tem
f
r
o
m
is
s
u
es
lik
e
co
m
m
u
n
icatio
n
d
is
r
u
p
tio
n
s
o
r
m
eter
f
ail
u
r
es,
we
ex
am
in
ed
a
m
o
d
e
r
ate
r
an
g
e
o
f
0
.
5
%
to
2
%
m
is
s
in
g
n
ess
.
Fo
r
ea
ch
d
ataset,
we
s
im
u
lated
b
o
th
r
a
n
d
o
m
m
is
s
in
g
n
ess
,
wh
er
e
d
ata
is
MCAR
,
an
d
m
is
s
in
g
n
ess
b
ased
o
n
s
p
ec
if
ic
cr
it
er
ia
(
MA
R
)
.
Fo
r
MA
R
,
in
th
e
f
in
an
ce
d
atas
et,
m
is
s
in
g
v
alu
es
wer
e
in
tr
o
d
u
ce
d
o
n
lo
w
-
v
o
lu
m
e
tr
ad
in
g
d
ay
s
b
ased
o
n
th
e
ass
u
m
p
tio
n
th
at
d
u
r
in
g
lo
w
-
v
o
lu
m
e
d
ay
s
,
s
to
ck
p
r
ice
d
ata
m
ig
h
t
b
e
in
c
o
m
p
lete
d
u
e
to
less
m
ar
k
et
ac
tiv
ity
.
Fo
r
th
e
h
ea
lth
ca
r
e
d
ataset,
th
e
m
is
s
in
g
h
ea
r
t
r
ate
v
alu
es
wer
e
in
tr
o
d
u
ce
d
d
u
r
in
g
p
e
r
io
d
s
o
f
h
ig
h
h
ea
r
t
r
ate.
Fo
r
th
e
en
er
g
y
d
ataset,
m
is
s
in
g
v
alu
es
wer
e
in
tr
o
d
u
c
ed
m
o
r
e
f
r
e
q
u
en
tly
d
u
r
in
g
p
e
ak
d
em
an
d
h
o
u
r
s
as
r
ea
l
-
wo
r
ld
en
er
g
y
s
y
s
tem
s
d
ata
co
llectio
n
er
r
o
r
s
m
ay
b
e
co
r
r
elate
d
with
h
ig
h
en
e
r
g
y
u
s
ag
e
p
er
i
o
d
s
.
T
h
is
allo
wed
u
s
to
in
v
esti
g
at
e
h
o
w
d
if
f
er
en
t
m
is
s
in
g
n
ess
m
ec
h
an
is
m
s
m
ig
h
t
af
f
ec
t
th
e
p
er
f
o
r
m
an
ce
o
f
o
u
r
f
o
r
ec
asti
n
g
m
o
d
els.
T
o
f
u
r
th
e
r
en
h
an
ce
th
e
co
m
p
r
eh
en
s
iv
e
n
ess
o
f
o
u
r
an
aly
s
is
,
we
em
p
lo
y
ed
th
r
ee
d
is
tin
ct
im
p
u
tatio
n
tech
n
iq
u
es:
f
o
r
wa
r
d
f
ill,
lin
e
ar
in
te
r
p
o
latio
n
,
a
n
d
k
-
NN
im
p
u
tatio
n
.
T
h
ese
tech
n
iq
u
es
r
e
p
r
esen
t
a
r
an
g
e
o
f
co
m
p
le
x
ity
in
h
a
n
d
lin
g
m
is
s
in
g
d
ata,
f
r
o
m
s
im
p
le
r
ep
lace
m
en
t
with
p
r
ev
i
o
u
s
v
alu
es
to
m
o
r
e
s
o
p
h
is
ticated
esti
m
atio
n
s
b
ase
d
o
n
n
eig
h
b
o
r
in
g
d
ata
p
o
in
ts
.
2
.
6
.
E
v
a
lua
t
i
o
n
m
et
rics
T
h
e
m
o
d
els we
r
e
e
v
alu
ated
o
n
th
eir
f
o
r
ec
asti
n
g
ac
cu
r
ac
y
u
s
in
g
th
e
f
o
llo
win
g
m
et
r
ics:
−
Me
an
ab
s
o
lu
te
er
r
o
r
(
MA
E
)
:
t
h
is
m
ea
s
u
r
es th
e
av
er
ag
e
m
ag
n
itu
d
e
o
f
er
r
o
r
s
in
th
e
p
r
ed
icti
o
n
s
.
−
Me
an
ab
s
o
lu
te
p
er
ce
n
ta
g
e
er
r
o
r
(
MA
PE)
:
th
is
m
ea
s
u
r
es
th
e
er
r
o
r
as
a
p
er
ce
n
tag
e
o
f
t
h
e
ac
tu
al
v
alu
e,
p
r
o
v
id
i
n
g
a
n
o
r
m
alize
d
m
ea
s
u
r
e
o
f
ac
c
u
r
ac
y
.
−
R
-
s
q
u
ar
ed
(
R
²)
:
th
is
q
u
an
tifie
s
h
o
w
well
th
e
m
o
d
el
ex
p
lain
s
t
h
e
v
ar
ian
ce
i
n
th
e
d
ataset.
T
ab
le
1
.
Hy
p
er
p
ar
a
m
eter
s
elec
tio
n
M
o
d
e
l
H
y
p
e
r
p
a
r
a
me
t
e
r
R
a
n
g
e
/
v
a
l
u
e
S
e
l
e
c
t
e
d
v
a
l
u
e
(
v
a
l
i
d
a
t
i
o
n
se
t
)
A
R
I
M
A
p
(
a
u
t
o
r
e
g
r
e
ss
i
v
e
t
e
r
m)
[
0
,
1
,
2
,
3
,
4
,
5
]
2
d
(
d
i
f
f
e
r
e
n
c
i
n
g
t
e
r
m)
[
0
,
1
,
2
]
1
q
(
m
o
v
i
n
g
a
v
e
r
a
g
e
t
e
r
m)
[
0
,
1
,
2
,
3
,
4
,
5
]
2
A
I
C
O
p
t
i
mi
z
e
d
-
LSTM
N
u
mb
e
r
o
f
LST
M
u
n
i
t
s
[
5
0
,
1
0
0
,
1
5
0
,
2
0
0
]
1
0
0
D
r
o
p
o
u
t
r
a
t
e
[
0
.
1
,
0
.
2
,
0
.
3
,
0
.
4
]
0
.
2
Le
a
r
n
i
n
g
r
a
t
e
[
0
.
0
0
1
,
0
.
0
0
5
,
0
.
0
1
]
0
.
0
0
1
B
a
t
c
h
si
z
e
[
1
6
,
3
2
,
6
4
]
32
N
u
mb
e
r
o
f
e
p
o
c
h
s
[
5
0
,
1
0
0
,
1
5
0
,
2
0
0
]
1
0
0
O
p
t
i
mi
z
e
r
A
d
a
m
A
d
a
m
TFT
N
u
mb
e
r
o
f
a
t
t
e
n
t
i
o
n
h
e
a
d
s
[
2
,
4
,
8
]
4
H
i
d
d
e
n
l
a
y
e
r
s
i
z
e
[
1
6
,
3
2
,
6
4
,
1
2
8
]
32
N
u
mb
e
r
o
f
G
R
N
l
a
y
e
r
s
[
1
,
2
,
3
]
2
D
r
o
p
o
u
t
r
a
t
e
[
0
.
1
,
0
.
2
,
0
.
3
,
0
.
4
,
0
.
5
]
0
.
2
Le
a
r
n
i
n
g
r
a
t
e
[
0
.
0
0
0
5
,
0
.
0
0
1
,
0
.
0
0
5
,
0
.
0
1
]
0
.
0
0
1
B
a
t
c
h
si
z
e
[
1
6
,
3
2
,
6
4
]
32
N
u
mb
e
r
o
f
e
p
o
c
h
s
[
5
0
,
1
0
0
,
1
5
0
,
2
0
0
]
1
0
0
O
p
t
i
mi
z
e
r
A
d
a
m
A
d
a
m
3.
RE
SU
L
T
S
3
.
1
.
F
ina
ncia
l da
t
a
s
et
re
s
ults
I
n
th
e
f
in
an
cial
d
o
m
ai
n
,
th
e
o
b
jectiv
e
was
to
f
o
r
ec
ast
th
e
clo
s
in
g
p
r
ice
o
f
th
e
S&
P
5
0
0
in
d
ex
u
s
in
g
h
is
to
r
ical
s
to
ck
p
r
ice
d
ata.
T
h
e
ev
alu
atio
n
m
etr
ics
f
o
r
AR
I
MA
,
L
STM
,
an
d
T
FT
m
o
d
els
o
n
th
e
test
s
et
ar
e
s
u
m
m
ar
ized
in
T
ab
le
2
.
T
h
e
AR
I
MA
m
o
d
el
s
h
o
wed
r
ea
s
o
n
ab
le
p
e
r
f
o
r
m
an
ce
in
p
r
ed
i
ctin
g
th
e
S&
P
5
0
0
in
d
ex
,
b
u
t
its
ab
ilit
y
to
ca
p
tu
r
e
n
o
n
-
lin
ea
r
p
atter
n
s
was
lim
ited
.
T
h
e
m
o
d
el
ac
h
iev
ed
a
MA
E
o
f
1
5
.
2
USD
an
d
a
MA
PE
o
f
1
.
9
%,
in
d
icatin
g
th
at
th
e
p
r
ed
ictio
n
s
wer
e
with
in
ap
p
r
o
x
im
ately
2
%
o
f
th
e
ac
tu
al
v
alu
es.
Ho
wev
er
,
th
e
R
²
v
alu
e
o
f
0
.
7
2
s
h
o
wed
th
at
th
e
m
o
d
el
ex
p
lain
ed
7
2
%
o
f
th
e
v
ar
ian
ce
in
th
e
d
ata.
T
h
e
L
STM
m
o
d
el,
d
u
e
to
its
ab
ilit
y
to
ca
p
t
u
r
e
lo
n
g
-
ter
m
d
ep
en
d
en
cies
a
n
d
n
o
n
-
lin
e
ar
r
e
latio
n
s
h
ip
s
,
s
ig
n
if
ican
tly
o
u
tp
er
f
o
r
m
ed
A
R
I
MA
.
I
t
ac
h
iev
ed
a
MA
E
o
f
1
0
.
1
USD
an
d
a
MA
PE
o
f
1
.
3
%.
T
h
e
R
²
v
alu
e
was
0
.
8
7
,
in
d
icatin
g
t
h
at
th
e
L
STM
m
o
d
el
ex
p
lain
ed
8
7
%
o
f
th
e
v
ar
ian
ce
in
th
e
s
to
ck
p
r
ic
es.
T
h
e
T
FT
m
o
d
el
d
em
o
n
s
tr
ated
th
e
b
est
p
er
f
o
r
m
an
ce
,
lev
e
r
ag
in
g
atten
tio
n
m
ec
h
an
is
m
s
to
ca
p
tu
r
e
b
o
th
s
h
o
r
t
-
an
d
lo
n
g
-
ter
m
d
ep
en
d
e
n
cies.
I
t
ac
h
iev
ed
th
e
lo
west
MA
E
o
f
8
.
6
USD
a
n
d
a
MA
PE
o
f
1
.
1
%,
with
an
R
²
o
f
0
.
9
1
,
s
h
o
win
g
th
at
th
e
m
o
d
el
was
ab
le
to
e
x
p
lain
9
1
%
o
f
th
e
v
ar
ian
ce
in
s
to
ck
p
r
ices,
o
u
tp
er
f
o
r
m
in
g
b
o
th
AR
I
MA
an
d
L
STM
.
Evaluation Warning : The document was created with Spire.PDF for Python.
C
o
m
p
u
t Sci
I
n
f
T
ec
h
n
o
l
I
SS
N:
2722
-
3
2
2
1
Time
s
er
ies fo
r
ec
a
s
tin
g
:
a
co
mp
a
r
a
tive
a
n
a
lysi
s
o
f A
R
I
MA,
LS
TM
,
a
n
d
TFT m
o
d
els
…
(
Ma
r
ya
m
Ho
s
s
ein
i
)
297
3
.
2
.
H
ea
lt
hca
re
da
t
a
s
et
re
s
ults
Fo
r
th
e
h
ea
lth
ca
r
e
d
ataset,
th
e
task
was
to
p
r
ed
ict
f
u
tu
r
e
h
ea
r
t
r
ate
v
alu
es
f
r
o
m
E
C
G
d
ata
d
u
r
in
g
s
leep
.
T
ab
le
2
p
r
esen
ts
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
AR
I
MA
,
L
STM
,
an
d
T
FT
m
o
d
els
o
n
th
e
test
s
et.
T
h
e
AR
I
MA
m
o
d
el
was
lim
ited
in
ca
p
tu
r
i
n
g
th
e
n
o
n
-
lin
ea
r
v
ar
iatio
n
s
in
h
ea
r
t
r
ate
p
atter
n
s
.
T
h
e
m
o
d
el
r
esu
lted
in
a
MA
E
o
f
6
.
3
b
p
m
an
d
a
MA
PE
o
f
4
.
7
%,
with
an
R
²
v
alu
e
o
f
0
.
6
8
.
T
h
is
in
d
icate
d
th
at
th
e
m
o
d
el
ex
p
lain
ed
6
8
%
o
f
th
e
v
ar
ian
ce
in
h
ea
r
t
r
ate
d
ata.
T
h
e
L
STM
m
o
d
el,
d
u
e
to
its
ab
ilit
y
to
h
an
d
le
s
eq
u
en
tial
an
d
n
o
n
-
li
n
e
ar
d
ata,
p
er
f
o
r
m
ed
b
etter
th
an
AR
I
MA
.
I
t
ac
h
iev
ed
a
MA
E
o
f
4
.
4
b
p
m
a
n
d
a
MA
PE
o
f
3
.
1
%,
with
an
R
²
o
f
0
.
8
1
.
T
h
e
T
FT
m
o
d
el
o
u
tp
er
f
o
r
m
ed
th
e
o
t
h
er
m
o
d
e
ls
in
f
o
r
ec
asti
n
g
h
ea
r
t
r
ate,
with
a
MA
E
o
f
3
.
8
b
p
m
an
d
a
MA
PE
o
f
2
.
9
%.
T
h
e
m
o
d
el
ac
h
iev
ed
an
R
²
v
al
u
e
o
f
0
.
8
5
,
d
em
o
n
s
tr
atin
g
th
at
it
ca
p
tu
r
ed
m
o
r
e
v
ar
ia
n
ce
in
th
e
h
ea
r
t
r
ate
tim
e
s
er
ies co
m
p
ar
ed
to
AR
I
MA
an
d
L
STM
.
3
.
3
.
E
nerg
y
da
t
a
s
et
re
s
ults
Fo
r
th
e
en
er
g
y
d
ataset,
th
e
g
o
al
was
to
p
r
ed
ict
h
o
u
r
ly
elec
tr
icity
d
em
an
d
.
T
h
e
p
er
f
o
r
m
a
n
ce
m
etr
ics
f
o
r
ea
ch
m
o
d
el
o
n
th
e
test
s
et
ar
e
p
r
o
v
id
e
d
in
T
ab
le
2
.
T
h
e
AR
I
MA
m
o
d
el
s
h
o
wed
d
ec
en
t
p
er
f
o
r
m
a
n
ce
,
with
a
MA
E
o
f
1
5
0
MWh
an
d
a
M
APE
o
f
2
.
7
%.
T
h
e
R
²
v
alu
e
o
f
0
.
7
6
in
d
icate
d
th
at
th
e
m
o
d
e
l
ex
p
lain
ed
7
6
%
o
f
th
e
v
ar
ian
ce
in
elec
tr
icity
d
em
an
d
.
T
h
e
L
STM
m
o
d
el
d
em
o
n
s
tr
at
ed
b
ette
r
p
er
f
o
r
m
a
n
ce
,
ac
h
iev
i
n
g
a
MA
E
o
f
1
1
0
MWh
an
d
a
MA
PE
o
f
1
.
9
%.
T
h
e
R
²
v
alu
e
was
0
.
8
5
,
s
h
o
win
g
a
h
ig
h
er
ab
ilit
y
to
c
ap
tu
r
e
p
atter
n
s
in
elec
tr
icity
d
em
an
d
co
m
p
ar
ed
t
o
AR
I
MA
.
T
h
e
T
FT
m
o
d
el
p
r
o
v
id
ed
th
e
b
est
r
esu
lts
,
with
a
MA
E
o
f
9
5
MWh
an
d
a
MA
PE
o
f
1
.
5
%.
Th
e
m
o
d
el
ac
h
iev
e
d
an
R
²
v
a
lu
e
o
f
0
.
8
9
,
in
d
icatin
g
th
at
it
ca
p
tu
r
ed
al
m
o
s
t
9
0
%
o
f
th
e
v
ar
ian
ce
in
t
h
e
d
ata,
o
u
tp
er
f
o
r
m
in
g
b
o
t
h
AR
I
MA
an
d
L
STM
.
T
ab
le
2
.
Mo
d
el
p
er
f
o
r
m
a
n
ce
o
n
f
in
an
cial,
h
ea
lth
ca
r
e,
a
n
d
en
er
g
y
d
atasets
(
test
s
et
r
esu
lts
)
M
o
d
e
l
D
a
t
a
s
e
t
M
A
E
M
A
P
E
(
%)
R²
A
R
I
M
A
F
i
n
a
n
c
i
a
l
(
U
S
D
)
1
5
.
2
1
.
9
0
.
7
2
LSTM
F
i
n
a
n
c
i
a
l
(
U
S
D
)
1
0
.
1
1
.
3
0
.
8
7
TFT
F
i
n
a
n
c
i
a
l
(
U
S
D
)
8
.
6
1
.
1
0
.
9
1
A
R
I
M
A
H
e
a
l
t
h
c
a
r
e
(
b
p
m)
6
.
3
4
.
7
0
.
6
8
LSTM
H
e
a
l
t
h
c
a
r
e
(
b
p
m)
4
.
4
3
.
1
0
.
8
1
TFT
H
e
a
l
t
h
c
a
r
e
(
b
p
m)
3
.
8
2
.
9
0
.
8
5
A
R
I
M
A
En
e
r
g
y
(
M
W
h
)
1
5
0
2
.
7
0
.
7
6
LSTM
En
e
r
g
y
(
M
W
h
)
1
1
0
1
.
9
0
.
8
5
TFT
En
e
r
g
y
(
M
W
h
)
95
1
.
5
0
.
8
9
3
.
4
.
O
v
er
a
ll
re
s
ults
Acr
o
s
s
all
d
atasets
,
th
e
T
FT
c
o
n
s
is
ten
tly
o
u
tp
er
f
o
r
m
ed
t
h
e
AR
I
MA
an
d
L
STM
m
o
d
els.
T
h
is
ca
n
b
e
attr
ib
u
ted
to
T
FT'
s
ab
ilit
y
to
ca
p
tu
r
e
b
o
th
s
h
o
r
t
-
an
d
lo
n
g
-
ter
m
d
ep
e
n
d
en
cies
in
th
e
d
ata,
as
well
as
it
s
ef
f
icien
t
u
s
e
o
f
atten
tio
n
m
ec
h
an
is
m
s
.
T
h
e
L
STM
m
o
d
el
s
h
o
wed
b
etter
p
e
r
f
o
r
m
an
ce
th
a
n
AR
I
MA
in
m
o
s
t
ca
s
es,
p
ar
ticu
lar
ly
wh
en
d
ea
lin
g
with
n
o
n
-
lin
ea
r
an
d
c
o
m
p
lex
tim
e
s
er
ies
d
ata,
s
u
c
h
as
h
ea
r
t
r
ate
an
d
elec
tr
icity
d
em
an
d
.
T
h
ese
r
esu
lts
s
u
g
g
est
th
at
wh
ile
AR
I
MA
r
em
ain
s
a
r
elia
b
le
m
o
d
el
f
o
r
s
im
p
ler
,
lin
ea
r
tim
e
s
er
ies,
d
ee
p
lear
n
in
g
m
o
d
els
s
u
ch
as
L
STM
an
d
T
FT
a
r
e
b
etter
s
u
ited
f
o
r
m
o
r
e
co
m
p
lex
p
atter
n
s
.
I
n
p
ar
ticu
lar
,
T
FT'
s
s
u
p
er
io
r
p
er
f
o
r
m
an
ce
h
ig
h
lig
h
ts
its
p
o
ten
tial
as
a
s
tate
-
of
-
th
e
-
a
r
t
s
o
lu
ti
o
n
f
o
r
tim
e
s
er
ies
f
o
r
ec
asti
n
g
in
v
ar
io
u
s
d
o
m
ain
s
.
3
.
5
.
Sens
it
iv
it
y
a
na
l
y
s
is
t
o
mis
s
i
ng
da
t
a
R
esu
lts
o
f
th
e
s
en
s
itiv
ity
an
a
ly
s
is
ac
r
o
s
s
all
th
r
ee
d
ataset
s
(
f
in
an
cial,
h
ea
lth
ca
r
e,
an
d
en
er
g
y
)
ar
e
p
r
esen
ted
in
Fig
u
r
e
1
.
T
h
ese
f
ig
u
r
es
d
etail
th
e
MA
PE
p
er
f
o
r
m
an
ce
m
etr
ics
f
o
r
th
e
AR
I
M
A,
L
STM
,
an
d
T
FT
m
o
d
els u
n
d
e
r
d
if
f
e
r
en
t m
is
s
in
g
d
ata
s
ce
n
ar
io
s
an
d
im
p
u
tatio
n
tech
n
iq
u
es.
As ex
p
ec
te
d
,
th
e
p
er
f
o
r
m
an
ce
o
f
all
m
o
d
els
d
eg
r
ad
es
as
th
e
p
er
c
en
tag
e
o
f
m
is
s
in
g
d
ata
in
cr
e
ases
.
Ho
we
v
er
,
th
e
ex
ten
t
o
f
d
eg
r
ad
atio
n
v
ar
ies
ac
r
o
s
s
m
o
d
els an
d
d
atasets
.
T
h
e
T
FT
m
o
d
el
co
n
s
is
ten
tly
d
em
o
n
s
tr
ates
g
r
ea
ter
r
o
b
u
s
tn
e
s
s
to
m
is
s
in
g
d
ata
co
m
p
ar
e
d
t
o
AR
I
MA
an
d
L
STM
ac
r
o
s
s
all
d
ata
s
ets,
p
ar
ticu
lar
ly
in
s
ce
n
ar
io
s
with
h
ig
h
er
lev
els
o
f
m
is
s
in
g
n
ess
.
T
h
is
s
u
g
g
ests
th
at
T
FT's
ab
ilit
y
to
ca
p
tu
r
e
co
m
p
lex
tem
p
o
r
al
d
ep
en
d
en
cies
an
d
u
tili
ze
atten
tio
n
m
ec
h
an
i
s
m
s
m
ak
es
it
le
s
s
s
u
s
ce
p
tib
le
to
th
e
n
eg
ativ
e
i
m
p
ac
t
o
f
m
is
s
in
g
d
ata.
T
h
e
ch
o
ice
o
f
im
p
u
tatio
n
tech
n
iq
u
e
in
f
lu
en
ce
s
m
o
d
el
p
er
f
o
r
m
an
ce
.
I
n
g
e
n
er
al,
k
-
NN
im
p
u
ta
tio
n
y
ield
s
b
etter
r
e
s
u
lts
th
an
f
o
r
war
d
f
ill
an
d
li
n
ea
r
in
ter
p
o
latio
n
,
esp
ec
ially
f
o
r
L
STM
an
d
T
FT
.
T
h
is
in
d
icate
s
th
at
m
o
r
e
s
o
p
h
is
ticated
im
p
u
tatio
n
tech
n
iq
u
es c
an
h
elp
m
itig
ate
th
e
ad
v
er
s
e
ef
f
ec
ts
o
f
m
is
s
in
g
d
ata
o
n
f
o
r
ec
asti
n
g
ac
cu
r
ac
y
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
7
2
2
-
3
2
2
1
C
o
m
p
u
t Sci
I
n
f
T
ec
h
n
o
l
,
Vo
l.
7
,
No
.
3
,
No
v
em
b
er
20
26
:
2
9
1
-
303
298
Fo
r
th
e
f
in
an
cia
l
d
ataset
as
s
h
o
wn
in
Fig
u
r
e
1
(
a)
,
th
e
m
o
d
els
ar
e
g
en
er
ally
r
o
b
u
s
t
to
lo
w
lev
els
o
f
m
is
s
in
g
d
ata
(
u
p
to
1
%).
As
m
is
s
in
g
n
ess
in
cr
ea
s
es,
AR
I
M
A's
p
er
f
o
r
m
a
n
ce
d
e
g
r
ad
es
m
o
r
e
s
ig
n
if
ican
tly
t
h
an
L
STM
an
d
T
FT.
Fo
r
th
e
h
ea
l
th
ca
r
e
d
ataset
,
in
Fig
u
r
e
1
(
b
)
,
T
FT
ex
h
ib
i
ts
th
e
g
r
ea
test
r
o
b
u
s
tn
ess
to
m
is
s
in
g
d
ata,
m
ain
tain
in
g
r
elativ
ely
g
o
o
d
p
e
r
f
o
r
m
an
ce
e
v
en
w
ith
2
0
%
m
is
s
in
g
n
ess
.
L
STM
s
h
o
ws
m
o
d
er
at
e
s
en
s
itiv
ity
,
wh
ile
AR
I
MA
is
t
h
e
m
o
s
t
af
f
ec
te
d
.
Fo
r
th
e
e
n
er
g
y
d
ataset
,
as
s
h
o
wn
in
Fig
u
r
e
1
(
c)
,
s
im
ilar
to
th
e
f
in
an
cial
d
ata
s
et,
th
e
m
o
d
els
ar
e
r
o
b
u
s
t
to
lo
w
lev
els
o
f
m
is
s
in
g
n
ess
.
As
m
is
s
in
g
n
ess
in
cr
ea
s
es,
AR
I
MA
'
s
p
er
f
o
r
m
an
ce
d
ec
lin
es m
o
r
e
r
a
p
id
ly
co
m
p
ar
ed
t
o
L
STM
an
d
T
FT.
(
a)
(
b
)
(
c)
Fig
u
r
e
1
.
Per
f
o
r
m
an
c
e
co
m
p
ar
is
o
n
o
f
AR
I
MA
,
L
STM
,
an
d
T
FT
m
o
d
els u
n
d
e
r
v
ar
y
in
g
m
is
s
in
g
-
d
ata
co
n
d
itio
n
s
an
d
im
p
u
tatio
n
tec
h
n
iq
u
es a
cr
o
s
s
th
r
ee
ap
p
licatio
n
d
o
m
ain
s
in
(
a)
f
in
an
cial
d
ataset,
(
b
)
h
ea
lth
ca
r
e
d
ataset,
an
d
(
c)
e
n
er
g
y
d
ataset
.
Fo
r
ec
asti
n
g
p
er
f
o
r
m
a
n
ce
is
ev
a
lu
ated
u
s
in
g
MA
PE
u
n
d
er
b
o
th
MCAR
an
d
MA
R
s
ce
n
ar
io
s
with
f
o
r
war
d
f
ill,
lin
ea
r
in
ter
p
o
latio
n
,
an
d
k
-
NN
im
p
u
tatio
n
.
Acr
o
s
s
all
d
at
asets
,
f
o
r
ec
asti
n
g
ac
cu
r
ac
y
g
e
n
er
ally
d
ec
r
ea
s
es a
s
m
is
s
in
g
n
ess
in
cr
ea
s
es
; h
o
wev
er
,
T
FT
d
em
o
n
s
tr
ates g
r
ea
te
r
r
o
b
u
s
tn
ess
u
n
d
er
in
c
r
ea
s
in
g
m
is
s
in
g
-
d
ata
lev
els
,
p
ar
ticu
lar
ly
u
n
d
e
r
MA
R
co
n
d
itio
n
s
an
d
h
i
g
h
er
m
is
s
in
g
n
ess
r
ates
Evaluation Warning : The document was created with Spire.PDF for Python.
C
o
m
p
u
t Sci
I
n
f
T
ec
h
n
o
l
I
SS
N:
2722
-
3
2
2
1
Time
s
er
ies fo
r
ec
a
s
tin
g
:
a
co
mp
a
r
a
tive
a
n
a
lysi
s
o
f A
R
I
MA,
LS
TM
,
a
n
d
TFT m
o
d
els
…
(
Ma
r
ya
m
Ho
s
s
ein
i
)
299
W
h
en
co
m
p
ar
in
g
th
e
im
p
ac
t
o
f
MCAR
v
er
s
u
s
MA
R
m
is
s
in
g
n
ess
m
ec
h
an
is
m
s
,
we
o
b
s
er
v
e
a
g
en
er
al
tr
en
d
wh
er
e
MA
R
s
ce
n
ar
io
s
l
ea
d
to
s
lig
h
tly
wo
r
s
e
m
o
d
el
p
er
f
o
r
m
a
n
ce
ac
r
o
s
s
all
d
atasets
.
T
h
is
s
u
g
g
ests
th
at
wh
en
m
is
s
in
g
v
al
u
es
ar
e
d
ep
en
d
en
t
o
n
o
b
s
er
v
e
d
d
ata,
th
e
im
p
u
tatio
n
p
r
o
ce
s
s
b
ec
o
m
es
m
o
r
e
ch
allen
g
i
n
g
,
p
o
ten
tially
lead
in
g
to
less
r
eli
ab
le
esti
m
ates.
T
h
e
T
FT
m
o
d
el
r
em
ain
s
th
e
m
o
s
t r
o
b
u
s
t u
n
d
er
MA
R
co
n
d
itio
n
s
,
r
ein
f
o
r
cin
g
its
ab
ilit
y
to
a
d
ap
t
to
m
o
r
e
co
m
p
le
x
m
is
s
in
g
n
e
s
s
s
tr
u
ctu
r
es.
C
o
n
v
er
s
ely
,
AR
I
MA
is
th
e
m
o
s
t
s
en
s
itiv
e
to
MA
R
,
with
p
er
f
o
r
m
an
ce
d
eter
io
r
atin
g
s
ig
n
if
ican
tly
as th
e
m
is
s
in
g
r
ate
in
cr
ea
s
es.
4.
DIS
CU
SS
I
O
N
T
im
e
s
er
ies
f
o
r
ec
asti
n
g
is
a
cr
u
cial
task
in
m
u
ltip
le
d
o
m
ain
s
,
in
clu
d
in
g
f
in
a
n
ce
,
h
ea
lth
ca
r
e,
a
n
d
en
er
g
y
.
Acc
u
r
ate
p
r
ed
ictio
n
s
o
f
f
u
t
u
r
e
tr
en
d
s
in
th
ese
ar
ea
s
en
ab
le
b
etter
d
ec
is
io
n
-
m
ak
in
g
,
r
is
k
m
an
ag
e
m
en
t,
an
d
o
p
e
r
atio
n
al
ef
f
icien
c
y
.
F
o
r
in
s
tan
ce
,
p
r
e
d
ictin
g
s
to
ck
p
r
ices
h
elp
s
in
v
esto
r
s
m
ak
e
in
f
o
r
m
e
d
tr
ad
in
g
d
ec
is
io
n
s
,
f
o
r
ec
asti
n
g
h
ea
r
t
r
a
te
ca
n
ass
is
t
in
d
etec
tin
g
an
o
m
alies
r
elate
d
to
s
leep
d
is
o
r
d
er
s
,
a
n
d
p
r
ed
ictin
g
en
er
g
y
d
em
a
n
d
is
v
ital
f
o
r
o
p
tim
izin
g
p
o
wer
g
r
id
s
.
Ho
w
ev
er
,
th
e
in
h
e
r
en
t
co
m
p
lex
it
y
o
f
th
ese
d
atasets
,
ch
ar
ac
ter
ized
b
y
n
o
n
-
lin
ea
r
it
y
,
s
ea
s
o
n
ality
,
a
n
d
lo
n
g
-
ter
m
d
ep
e
n
d
en
cies
,
m
ak
es
f
o
r
ec
a
s
tin
g
a
ch
allen
g
in
g
task
.
T
r
ad
itio
n
al
lin
ea
r
m
o
d
els
li
k
e
AR
I
MA
o
f
ten
f
all
s
h
o
r
t
i
n
ca
p
tu
r
in
g
th
ese
in
tr
icate
p
atter
n
s
,
n
ec
ess
itatin
g
th
e
u
s
e
o
f
m
o
r
e
ad
v
an
ce
d
d
ee
p
lear
n
in
g
m
eth
o
d
s
.
4
.
1
.
Su
mm
a
ry
o
f
f
ind
ing
s
I
n
th
is
s
tu
d
y
,
we
co
m
p
ar
ed
t
h
e
p
er
f
o
r
m
an
ce
o
f
th
r
ee
f
o
r
ec
a
s
tin
g
m
o
d
els:
AR
I
MA
,
L
STM
,
an
d
T
FT,
ac
r
o
s
s
th
r
ee
d
iv
er
s
e
d
atasets
(
f
in
an
cial,
h
ea
lth
ca
r
e,
a
n
d
e
n
er
g
y
)
.
Ou
r
f
in
d
in
g
s
s
h
o
w
t
h
at
wh
ile
AR
I
MA
p
er
f
o
r
m
ed
a
d
eq
u
ately
in
s
im
p
ler
ca
s
es,
it
was
s
ig
n
if
ican
t
ly
o
u
tp
e
r
f
o
r
m
ed
b
y
th
e
d
ee
p
lear
n
in
g
m
o
d
els
(
L
STM
an
d
T
FT)
,
p
ar
ticu
lar
l
y
in
d
atasets
w
ith
n
o
n
-
lin
ea
r
d
ep
en
d
en
cies.
T
h
e
AR
I
MA
m
o
d
el
y
ield
ed
th
e
h
ig
h
est
er
r
o
r
s
an
d
l
o
west
ex
p
lain
ed
v
ar
ian
ce
,
with
MA
E
v
alu
es
o
f
1
5
.
2
USD
f
o
r
th
e
f
in
an
cial
d
ataset,
6
.
3
b
p
m
f
o
r
h
ea
lth
ca
r
e,
an
d
1
5
0
MWh
f
o
r
e
n
er
g
y
,
alo
n
g
s
id
e
r
elativ
ely
lo
w
R
²
v
alu
es.
L
STM
,
as
ex
p
ec
ted
,
o
u
tp
e
r
f
o
r
m
ed
AR
I
MA
with
s
ig
n
if
ican
tly
lo
wer
er
r
o
r
s
ac
r
o
s
s
all
d
ataset
s
,
d
em
o
n
s
tr
atin
g
its
s
tr
en
g
th
in
h
an
d
lin
g
s
eq
u
e
n
tial
an
d
n
o
n
-
l
in
ea
r
d
ata.
Fo
r
ex
am
p
le
,
L
STM
r
ed
u
ce
d
th
e
MA
E
to
1
0
.
1
USD
f
o
r
f
in
an
cial
d
ata
an
d
4
.
4
b
p
m
f
o
r
h
ea
lth
ca
r
e.
Fin
ally
,
th
e
T
FT
m
o
d
el
d
e
liv
er
e
d
th
e
b
est
p
er
f
o
r
m
a
n
ce
,
d
e
m
o
n
s
tr
atin
g
s
u
p
er
io
r
ac
c
u
r
ac
y
an
d
th
e
ab
il
ity
to
ca
p
tu
r
e
b
o
th
s
h
o
r
t
-
an
d
l
o
n
g
-
ter
m
d
e
p
en
d
e
n
cies.
T
FT
ac
h
iev
ed
t
h
e
lo
west
MA
E
o
f
8
.
6
USD
f
o
r
f
in
an
c
ial
d
ata,
3
.
8
b
p
m
f
o
r
h
ea
lth
ca
r
e,
an
d
9
5
MWh
f
o
r
e
n
er
g
y
,
with
R
²
v
alu
es
ex
ce
ed
in
g
0
.
8
5
ac
r
o
s
s
all
d
o
m
ain
s
.
4
.
2
.
Co
m
pa
riso
n wit
h
lite
ra
t
ure
Ou
r
f
in
d
in
g
s
alig
n
with
p
r
e
v
io
u
s
r
esear
ch
,
wh
ich
co
n
s
i
s
ten
tly
d
em
o
n
s
tr
ates
th
e
lim
itatio
n
s
o
f
AR
I
MA
in
d
ea
lin
g
with
n
o
n
-
lin
ea
r
an
d
co
m
p
lex
tim
e
s
er
ies
.
Stu
d
ies
co
m
p
a
r
in
g
AR
I
MA
with
d
ee
p
lear
n
in
g
m
o
d
els
in
s
to
ck
m
ar
k
et
p
r
e
d
ictio
n
s
,
s
u
ch
as
th
o
s
e
b
y
Siam
i
-
Nam
in
i
et
a
l
.
[
2
4
]
,
[
25]
,
r
ep
o
r
t
s
im
ilar
p
er
f
o
r
m
an
ce
g
a
p
s
,
with
d
ee
p
lear
n
in
g
m
o
d
els
lik
e
L
STM
an
d
B
iLST
M
r
ed
u
cin
g
e
r
r
o
r
r
a
tes
b
y
8
0
%
to
9
0
%
co
m
p
ar
ed
t
o
AR
I
MA
.
I
n
an
o
th
er
s
tu
d
y
,
Siam
i
-
Nam
in
i
et
a
l
.
[
2
6
]
d
em
o
n
s
tr
ated
th
at
L
STM
s
ig
n
if
ican
tly
o
u
tp
er
f
o
r
m
ed
AR
I
MA
in
f
in
an
cial
tim
e
s
er
ies
f
o
r
ec
asti
n
g
,
r
ed
u
cin
g
er
r
o
r
s
b
y
ab
o
u
t
8
0
%
ac
r
o
s
s
v
ar
io
u
s
f
in
an
cial
d
atasets
,
in
clu
d
in
g
s
to
ck
in
d
ices
a
n
d
ec
o
n
o
m
ic
i
n
d
icato
r
s
.
I
n
t
h
e
h
ea
lth
ca
r
e
d
o
m
ain
,
L
STM
h
as
b
ee
n
s
h
o
wn
to
o
u
t
p
er
f
o
r
m
tr
a
d
itio
n
al
m
o
d
els
in
class
if
y
in
g
p
h
y
s
io
lo
g
ical
s
ig
n
als
lik
e
E
C
G,
as
d
em
o
n
s
tr
ated
in
th
e
wo
r
k
b
y
Yild
ir
im
[
3
]
,
wh
er
e
an
L
STM
-
b
ased
m
o
d
el
ac
h
i
ev
ed
o
v
er
9
9
%
ac
cu
r
ac
y
in
class
if
y
in
g
E
C
G
s
ig
n
als,
s
ig
n
if
ican
tly
o
u
tp
er
f
o
r
m
in
g
tr
ad
itio
n
al
m
eth
o
d
s
s
u
ch
as
s
u
p
p
o
r
t v
ec
t
o
r
m
ac
h
in
es (
SVM)
.
T
h
e
T
FT,
a
r
elativ
ely
r
ec
e
n
t
ad
d
itio
n
to
tim
e
s
er
ies
f
o
r
ec
asti
n
g
,
h
as
d
em
o
n
s
tr
ate
d
s
u
p
er
io
r
p
er
f
o
r
m
an
ce
ac
r
o
s
s
m
u
ltip
l
e
d
o
m
ain
s
.
L
im
et
a
l
.
[
2
2
]
s
h
o
wed
th
at
T
FT
o
u
tp
e
r
f
o
r
m
s
tr
a
d
itio
n
al
m
o
d
els
an
d
ad
v
an
ce
d
r
ec
u
r
r
e
n
t
ar
ch
itectu
r
es
lik
e
L
STM
,
p
ar
ticu
lar
ly
in
en
er
g
y
d
em
an
d
f
o
r
ec
asti
n
g
an
d
m
u
lti
-
h
o
r
izo
n
s
ales
p
r
ed
ictio
n
,
d
u
e
to
its
ab
ilit
y
to
ca
p
tu
r
e
b
o
th
s
h
o
r
t
-
a
n
d
l
o
n
g
-
ter
m
d
ep
en
d
en
cies.
Similar
ly
,
N
iu
et
a
l.
[
2
7
]
co
n
f
ir
m
ed
T
F
T
'
s
ex
ce
p
tio
n
al
p
er
f
o
r
m
a
n
ce
i
n
win
d
p
o
we
r
f
o
r
ec
asti
n
g
,
w
h
er
e
it
s
ig
n
if
ican
tly
out
p
er
f
o
r
m
ed
L
STM
an
d
tr
ad
i
tio
n
al
m
o
d
els,
f
u
r
th
er
p
r
o
v
in
g
its
ef
f
ec
tiv
en
ess
in
h
an
d
lin
g
co
m
p
lex
,
n
o
n
-
lin
ea
r
r
elatio
n
s
h
ip
s
in
en
er
g
y
d
atasets
.
Ou
r
r
esu
lts
,
p
ar
ticu
lar
ly
th
e
im
p
r
o
v
e
d
ac
cu
r
ac
y
an
d
lo
wer
MA
E
v
alu
es
in
all
d
atasets
,
r
ein
f
o
r
ce
th
e
ad
v
a
n
ta
g
es
o
f
T
FT
in
d
ea
lin
g
with
ti
m
e
s
er
ies
co
n
tain
i
n
g
lo
n
g
-
ter
m
d
ep
e
n
d
en
cies
an
d
co
m
p
lex
p
atter
n
s
.
4
.
3
.
Str
eng
t
hs
,
lim
it
a
t
io
ns
,
a
nd
f
uture
direct
io
ns
T
h
is
s
tu
d
y
p
r
esen
ts
s
ev
er
al
s
tr
en
g
th
s
,
m
o
s
t
n
o
tab
ly
it
s
co
m
p
r
eh
en
s
iv
e
c
o
m
p
ar
is
o
n
o
f
th
r
ee
f
o
r
ec
asti
n
g
m
o
d
els
(
AR
I
MA
,
L
STM
,
an
d
T
FT
)
ac
r
o
s
s
d
i
s
tin
ct
d
o
m
ain
s
s
u
ch
as
f
in
a
n
ce
,
h
ea
lth
ca
r
e,
an
d
en
er
g
y
.
B
y
test
in
g
th
ese
m
o
d
els
o
n
d
iv
er
s
e
r
ea
l
-
wo
r
l
d
d
atasets
,
we
p
r
o
v
id
e
v
alu
ab
le
in
s
ig
h
ts
in
to
h
o
w
tr
ad
itio
n
al
an
d
a
d
v
an
ce
d
d
ee
p
lear
n
in
g
ap
p
r
o
ac
h
es
p
er
f
o
r
m
in
d
if
f
e
r
en
t
c
o
n
tex
ts
.
I
n
p
ar
tic
u
lar
,
th
e
in
clu
s
io
n
o
f
th
e
T
FT
m
o
d
el,
wh
ich
is
r
e
lativ
ely
n
ew
an
d
h
as
b
ee
n
less
f
r
eq
u
en
tly
ap
p
lied
in
s
o
m
e
a
r
ea
s
,
d
em
o
n
s
tr
ates
its
ab
ilit
y
to
o
u
tp
er
f
o
r
m
b
o
th
AR
I
MA
an
d
L
STM
in
ca
p
tu
r
in
g
co
m
p
lex
tim
e
s
er
ies
p
atter
n
s
.
T
h
e
r
ig
o
r
o
u
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
7
2
2
-
3
2
2
1
C
o
m
p
u
t Sci
I
n
f
T
ec
h
n
o
l
,
Vo
l.
7
,
No
.
3
,
No
v
em
b
er
20
26
:
2
9
1
-
303
300
m
eth
o
d
o
l
o
g
y
,
in
clu
d
i
n
g
t
h
e
u
s
e
o
f
r
o
llin
g
win
d
o
w
cr
o
s
s
-
v
ali
d
atio
n
,
e
n
s
u
r
es
th
at
f
u
tu
r
e
d
at
a
is
n
o
t
leak
e
d
in
t
o
th
e
tr
ain
in
g
p
r
o
ce
s
s
,
th
u
s
o
f
f
er
in
g
a
r
ea
lis
tic
ass
ess
m
en
t
o
f
m
o
d
el
p
e
r
f
o
r
m
an
ce
th
at
is
h
ig
h
ly
r
ele
v
an
t
f
o
r
p
r
ac
tical
ap
p
licatio
n
s
wh
er
e
f
o
r
ec
asti
n
g
ac
cu
r
ac
y
is
cr
it
ic
al.
I
n
ad
d
itio
n
,
th
e
im
p
ac
t
o
f
m
is
s
in
g
d
ata
an
d
im
p
u
tatio
n
tech
n
iq
u
es wa
s
s
tu
d
ied
.
B
ey
o
n
d
f
o
r
ec
asti
n
g
ac
c
u
r
ac
y
,
th
e
o
b
s
er
v
ed
r
o
b
u
s
tn
ess
o
f
T
FT
u
n
d
er
in
c
o
m
p
lete
d
ata
co
n
d
itio
n
s
m
ay
h
av
e
im
p
o
r
tan
t
im
p
licatio
n
s
f
o
r
lar
g
e
-
s
ca
le
in
f
o
r
m
atio
n
a
n
d
co
m
m
u
n
icatio
n
tech
n
o
lo
g
y
(
I
C
T
)
s
y
s
tem
s
an
d
en
ter
p
r
is
e
ap
p
licatio
n
s
.
I
n
p
r
ac
tical
en
v
ir
o
n
m
e
n
ts
,
tim
e
-
s
er
ies
d
ata
ar
e
in
cr
ea
s
in
g
ly
g
en
er
ated
th
r
o
u
g
h
d
is
tr
ib
u
ted
s
en
s
o
r
s
,
in
ter
n
et
-
of
-
th
in
g
s
(
I
o
T
)
in
f
r
astru
ctu
r
es,
clo
u
d
s
er
v
ices,
an
d
b
ig
-
d
ata
p
latf
o
r
m
s
,
w
h
er
e
m
is
s
in
g
v
alu
es
f
r
e
q
u
en
tly
ar
i
s
e
b
ec
au
s
e
o
f
c
o
m
m
u
n
icatio
n
f
ailu
r
es,
laten
cy
,
s
y
n
ch
r
o
n
i
za
tio
n
is
s
u
es,
an
d
s
en
s
o
r
in
ter
r
u
p
tio
n
s
.
T
h
e
ab
il
ity
o
f
T
FT
to
m
ain
tain
p
er
f
o
r
m
an
ce
u
n
d
er
m
is
s
in
g
-
d
ata
c
o
n
d
itio
n
s
s
u
g
g
ests
s
tr
o
n
g
p
o
ten
tial
f
o
r
d
e
p
lo
y
m
e
n
t
with
in
clo
u
d
-
b
ased
an
aly
ti
cs
p
ip
elin
es
an
d
lar
g
e
-
s
ca
le
f
o
r
ec
asti
n
g
s
y
s
tem
s
.
I
n
teg
r
atio
n
with
d
is
tr
ib
u
ted
c
o
m
p
u
tin
g
f
r
am
ewo
r
k
s
an
d
b
ig
-
d
ata
p
latf
o
r
m
s
c
o
u
ld
e
n
ab
le
r
e
al
-
tim
e
f
o
r
ec
asti
n
g
ap
p
licatio
n
s
in
s
m
ar
t
h
ea
lth
ca
r
e
s
y
s
tem
s
,
f
in
an
cial
m
o
n
ito
r
i
n
g
en
v
ir
o
n
m
en
ts
,
in
tellig
en
t
e
n
er
g
y
m
an
a
g
e
m
en
t
p
latf
o
r
m
s
,
an
d
e
n
ter
p
r
is
e
d
ec
i
s
io
n
-
s
u
p
p
o
r
t
s
y
s
tem
s
.
T
h
is
ca
p
ab
ilit
y
in
cr
ea
s
es
th
e
p
r
ac
tical
r
elev
an
ce
o
f
T
FT
b
ey
o
n
d
ex
p
er
im
en
tal
s
ettin
g
s
an
d
h
ig
h
lig
h
ts
its
p
o
ten
tial r
o
l
e
in
s
ca
lab
le
in
d
u
s
tr
ial
f
o
r
ec
asti
n
g
s
o
lu
tio
n
s
.
Desp
ite
th
ese
s
tr
en
g
th
s
,
th
e
s
t
u
d
y
h
as
ce
r
tain
lim
itatio
n
s
.
AR
I
MA
,
wh
ile
in
clu
d
ed
f
o
r
co
m
p
ar
is
o
n
,
is
s
o
m
ewh
at
o
u
td
ated
co
m
p
a
r
ed
to
m
o
d
e
r
n
d
ee
p
lear
n
in
g
m
o
d
els
an
d
its
lo
wer
p
er
f
o
r
m
an
ce
was
ex
p
ec
ted
.
L
STM
,
th
o
u
g
h
ef
f
ec
tiv
e,
lac
k
s
th
e
f
lex
ib
ilit
y
an
d
ca
p
ac
ity
o
f
T
FT,
esp
ec
ially
in
ca
p
tu
r
in
g
lo
n
g
-
r
a
n
g
e
d
ep
en
d
e
n
cies
ef
f
icien
tly
.
Fu
r
t
h
er
m
o
r
e
,
th
e
co
m
p
u
tatio
n
al
co
s
t
o
f
d
ee
p
lear
n
in
g
m
o
d
els,
p
ar
ticu
lar
ly
L
STM
an
d
T
FT,
is
a
n
o
tab
le
d
r
a
wb
ac
k
,
as
th
e
y
r
e
q
u
ir
e
s
ig
n
if
ican
t
r
eso
u
r
ce
s
,
esp
ec
ially
wh
en
f
in
e
-
t
u
n
in
g
h
y
p
er
p
ar
am
eter
s
th
r
o
u
g
h
m
eth
o
d
s
lik
e
g
r
id
s
ea
r
ch
.
L
o
o
k
in
g
ah
ea
d
,
th
er
e
ar
e
s
ev
er
al
av
en
u
es
f
o
r
f
u
tu
r
e
r
esea
r
ch
.
Alth
o
u
g
h
th
e
d
atasets
u
s
ed
in
th
is
s
tu
d
y
s
p
an
m
u
ltip
le
d
o
m
ain
s
an
d
p
r
o
v
id
e
a
b
r
o
ad
e
v
alu
atio
n
f
r
am
ewo
r
k
,
th
ey
p
r
im
ar
ily
co
n
s
is
t
o
f
b
en
ch
m
ar
k
d
atasets
co
m
m
o
n
ly
u
s
ed
f
o
r
a
lg
o
r
ith
m
ic
ev
a
l
u
atio
n
.
W
h
ile
b
en
ch
m
ar
k
d
atasets
f
ac
ilit
ate
r
ep
r
o
d
u
cib
ilit
y
an
d
s
tan
d
ar
d
ized
co
m
p
ar
is
o
n
,
t
h
ey
m
ay
n
o
t
f
u
lly
r
e
p
r
esen
t
th
e
co
m
p
lex
ity
a
n
d
v
ar
iab
i
lity
en
co
u
n
ter
e
d
in
o
p
er
atio
n
al
en
v
ir
o
n
m
en
ts
.
R
e
al
-
wo
r
ld
tim
e
-
s
er
ies
s
y
s
tem
s
f
r
eq
u
en
tly
ex
h
i
b
it
d
o
m
ai
n
-
s
p
e
cif
ic
ch
ar
ac
ter
is
tics
s
u
ch
as ir
r
eg
u
lar
s
am
p
lin
g
f
r
e
q
u
en
cies,
ev
o
lv
in
g
d
ata
d
is
tr
ib
u
tio
n
s
,
h
eter
o
g
e
n
eo
u
s
s
en
s
o
r
s
o
u
r
ce
s
,
o
p
er
atio
n
al
d
is
r
u
p
tio
n
s
,
a
n
d
v
ar
y
in
g
m
is
s
in
g
-
d
ata
p
atter
n
s
.
T
h
er
ef
o
r
e
,
th
e
g
en
e
r
aliza
b
ilit
y
o
f
th
e
p
r
o
p
o
s
ed
f
in
d
in
g
s
s
h
o
u
ld
b
e
in
ter
p
r
eted
w
ith
ca
u
tio
n
.
Fu
tu
r
e
s
tu
d
ies
s
h
o
u
ld
v
alid
ate
th
ese
f
o
r
ec
asti
n
g
ap
p
r
o
ac
h
es
u
s
in
g
s
ec
to
r
-
s
p
ec
if
ic
r
ea
l
-
wo
r
ld
d
atasets
f
r
o
m
in
d
u
s
tr
ial
en
v
ir
o
n
m
en
ts
,
h
ea
lth
ca
r
e
m
o
n
ito
r
in
g
s
y
s
tem
s
,
s
m
ar
t
g
r
id
s
,
f
in
an
cial
in
s
titu
tio
n
s
,
an
d
l
ar
g
e
-
s
ca
le
en
ter
p
r
is
e
p
latf
o
r
m
s
to
b
e
tter
ass
ess
r
o
b
u
s
tn
ess
an
d
p
r
ac
tical
d
ep
lo
y
m
e
n
t
p
o
ten
tial
u
n
d
e
r
r
ea
l
o
p
er
atio
n
al
co
n
d
itio
n
s
.
Mo
r
eo
v
er
,
f
u
t
u
r
e
wo
r
k
co
u
ld
ex
p
lo
r
e
th
e
g
en
er
aliza
tio
n
o
f
th
ese
m
o
d
el
s
to
m
o
r
e
ch
allen
g
in
g
s
ce
n
ar
i
o
s
,
s
u
ch
as
h
ig
h
ly
im
b
alan
c
e
d
o
r
n
o
is
y
d
atasets
,
wh
ich
r
em
ain
d
if
f
icu
lt f
o
r
d
ee
p
lear
n
in
g
ap
p
r
o
ac
h
es
[
2
8
]
.
W
h
ile
th
is
s
tu
d
y
f
o
cu
s
ed
o
n
AR
I
MA
,
L
STM
,
an
d
T
FT
m
o
d
els,
f
u
tu
r
e
r
esear
ch
c
o
u
l
d
ex
p
lo
r
e
a
wid
er
r
an
g
e
o
f
tim
e
s
er
ies
f
o
r
ec
asti
n
g
m
o
d
els.
Fo
r
AR
I
MA
,
th
is
co
u
ld
in
v
o
lv
e
in
v
esti
g
atin
g
v
ar
iatio
n
s
lik
e
SAR
I
MA
f
o
r
s
ea
s
o
n
ality
an
d
AR
I
MA
X
f
o
r
in
co
r
p
o
r
atin
g
e
x
ter
n
al
f
ac
to
r
s
.
Ad
d
itio
n
ally
,
co
m
p
ar
in
g
AR
I
MA
to
o
th
er
tr
ad
itio
n
al
m
o
d
els
li
k
e
E
x
p
o
n
en
tial
Sm
o
o
th
in
g
an
d
Ho
lt
-
W
in
ter
s
co
u
ld
p
r
o
v
id
e
v
alu
ab
le
in
s
ig
h
ts
.
I
n
th
e
co
n
tex
t
o
f
d
ee
p
lear
n
in
g
,
ex
p
lo
r
in
g
d
if
f
e
r
en
t
L
ST
M
v
ar
iatio
n
s
,
s
u
ch
as
B
i
-
L
STM
an
d
GR
U,
co
u
ld
f
u
r
th
er
e
n
h
an
ce
p
er
f
o
r
m
an
ce
.
C
o
m
p
ar
in
g
L
STM
to
o
th
e
r
d
e
ep
lear
n
in
g
m
o
d
els
lik
e
C
NNs
an
d
T
C
Ns
co
u
ld
also
r
ev
ea
l
th
eir
r
elativ
e
s
tr
e
n
g
th
s
an
d
wea
k
n
ess
es
f
o
r
tim
e
s
er
ies
f
o
r
ec
asti
n
g
.
Fo
r
tr
an
s
f
o
r
m
er
m
o
d
els,
in
v
esti
g
atin
g
v
ar
iatio
n
s
lik
e
t
r
an
s
f
o
r
m
er
-
XL
an
d
lo
n
g
f
o
r
m
er
co
u
ld
im
p
r
o
v
e
e
f
f
icien
cy
a
n
d
ad
d
r
ess
s
p
ec
if
ic
ch
allen
g
es.
C
o
m
p
ar
in
g
tr
an
s
f
o
r
m
er
s
to
o
th
er
atten
tio
n
-
b
ased
m
o
d
els
lik
e
I
n
f
o
r
m
er
co
u
ld
also
b
e
b
en
ef
icial.
I
t
is
im
p
o
r
tan
t
to
n
o
te
th
at
o
u
r
im
p
lem
en
tatio
n
o
f
T
FT
was
s
o
m
ewh
at
lim
ited
in
its
u
s
e
o
f
s
tatic
an
d
k
n
o
wn
f
u
tu
r
e
in
p
u
ts
.
W
h
ile
th
e
f
in
an
ce
d
ataset
r
elied
s
o
lely
o
n
o
b
s
er
v
ed
h
is
to
r
ical
in
p
u
ts
,
th
e
h
e
alth
ca
r
e
an
d
e
n
er
g
y
d
atas
ets
in
co
r
p
o
r
ate
d
lim
ited
k
n
o
wn
f
u
tu
r
e
co
v
ar
iates,
s
u
ch
as
tim
e
-
b
ased
f
ea
tu
r
es.
H
o
wev
er
,
we
d
id
n
o
t
f
u
lly
lev
er
ag
e
th
e
p
o
ten
tial
o
f
s
tatic
v
ar
iab
les
o
r
b
r
o
ad
er
k
n
o
wn
f
u
tu
r
e
in
p
u
ts
,
wh
ich
co
u
ld
en
h
an
ce
T
FT’
s
in
ter
p
r
etab
ilit
y
an
d
p
e
r
f
o
r
m
a
n
ce
.
Fu
tu
r
e
wo
r
k
co
u
l
d
in
v
es
tig
ate
th
e
im
p
ac
t
o
f
in
c
o
r
p
o
r
atin
g
a
r
ich
er
s
et
o
f
th
ese
in
p
u
ts
o
n
f
o
r
ec
asti
n
g
ac
c
u
r
ac
y
.
An
o
th
er
im
p
o
r
tan
t
d
ir
ec
tio
n
is
ad
d
r
ess
in
g
th
e
lack
o
f
in
ter
p
r
etab
ilit
y
in
m
o
d
els
lik
e
L
STM
an
d
T
FT.
W
h
ile
th
ese
m
o
d
els
p
er
f
o
r
m
ex
ce
p
tio
n
ally
well,
th
eir
“
b
lack
-
b
o
x
”
n
atu
r
e
m
ak
es
it
d
if
f
icu
lt
f
o
r
p
r
ac
titi
o
n
er
s
to
u
n
d
e
r
s
tan
d
th
e
u
n
d
er
ly
in
g
m
ec
h
an
is
m
s
d
r
iv
in
g
th
e
p
r
ed
i
ctio
n
s
[
2
2
]
.
Fu
tu
r
e
wo
r
k
co
u
ld
f
o
cu
s
o
n
en
h
an
cin
g
m
o
d
el
in
ter
p
r
etab
ilit
y
,
p
o
ten
ti
ally
b
y
in
co
r
p
o
r
atin
g
ex
p
lain
ab
ilit
y
tech
n
iq
u
es
th
at
wo
u
ld
m
ak
e
th
ese
m
o
d
els
m
o
r
e
tr
an
s
p
a
r
en
t a
n
d
ac
ce
s
s
ib
le,
esp
ec
ially
in
s
en
s
itiv
e
f
ield
s
lik
e
h
ea
lth
ca
r
e
[
2
9
]
.
Fin
ally
,
ex
p
lo
r
i
n
g
en
s
em
b
l
e
m
eth
o
d
s
th
at
co
m
b
in
e
t
h
e
s
tr
en
g
th
s
o
f
AR
I
MA
,
L
STM
,
a
n
d
T
FT
co
u
ld
o
f
f
er
a
p
r
o
m
is
in
g
p
ath
f
o
r
wa
r
d
[
3
0
]
.
Hy
b
r
id
m
o
d
els
th
at
u
t
ilize
AR
I
MA
’
s
lin
ea
r
ity
f
o
r
s
im
p
ler
p
atter
n
s
a
n
d
lev
er
ag
e
d
ee
p
lear
n
in
g
m
o
d
el
s
'
ab
ilit
y
to
ca
p
tu
r
e
c
o
m
p
lex
n
o
n
-
lin
ea
r
tr
en
d
s
co
u
l
d
y
ield
ev
en
b
etter
r
esu
lts
.
T
r
an
s
f
er
lear
n
in
g
an
d
d
o
m
ai
n
ad
ap
tatio
n
ar
e
also
ar
ea
s
wo
r
th
ex
p
l
o
r
in
g
,
esp
ec
ial
ly
f
o
r
ap
p
licatio
n
s
wh
er
e
lab
eled
d
ata
is
s
ca
r
ce
o
r
e
x
p
en
s
iv
e
to
co
llect
[
3
1
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.