I
AE
S
I
n
t
e
r
n
at
ion
al
Jou
r
n
al
of
Ar
t
if
icial
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
AI
)
Vol.
14,
No.
5:
Oc
tober
2025
,
pp.
4342
~
4352
I
S
S
N:
2252
-
8938
,
DO
I
:
10
.
11591/i
jai
.
v
14
.i
5
.
pp
43
42
-
4352
4342
Jou
r
n
al
h
omepage
:
ht
tp:
//
ij
ai
.
iaes
c
or
e
.
c
om
M
u
lti
-
c
la
ss s
t
oc
k
m
ar
k
e
t
f
o
r
e
c
ast
in
g w
it
h
d
e
e
p
l
e
a
r
n
in
g
m
o
d
e
ls
:
a
n
e
x
p
la
in
ab
le
a
r
t
ifi
c
ia
l
in
t
e
ll
ig
e
n
c
e
Chh
aya
P
a
t
e
l
1
,
As
h
win
Raiyan
i
2
1
D
e
pa
r
tm
e
nt
of
C
omput
e
r
E
ngi
ne
e
r
in
g,
S
c
hool
of
E
ngi
ne
e
r
in
g
,
RK
U
ni
ve
r
s
it
y,
R
a
jk
ot
,
I
ndi
a
2
I
ns
ti
tu
te
of
M
a
na
ge
me
nt
,
N
ir
ma
U
ni
ve
r
s
it
y,
A
hme
d
a
ba
d,
I
ndi
a
Ar
t
icle
I
n
f
o
AB
S
T
RA
CT
A
r
ti
c
le
h
is
tor
y
:
R
e
c
e
ived
M
a
r
12,
2025
R
e
vis
e
d
Aug
2,
2025
Ac
c
e
pted
S
e
p
7,
2025
In
t
h
i
s
res
earc
h
,
w
e
i
n
v
es
t
i
g
at
e
d
t
h
e
i
n
f
l
u
e
n
ce
o
f
d
i
f
feren
t
d
eep
l
earn
i
n
g
t
ech
n
i
q
u
e
s
o
n
t
i
me
s
er
i
es
s
t
o
c
k
mark
et
d
a
t
a,
es
p
eci
al
l
y
fo
r
al
l
N
i
f
t
y
5
0
co
mp
a
n
i
e
s
i
n
t
h
e
I
n
d
i
an
s
t
o
ck
mark
e
t
.
O
u
r
p
ro
p
o
s
ed
me
t
h
o
d
o
f
s
t
o
ck
mar
k
e
t
p
red
i
ct
i
o
n
fo
c
u
s
e
d
o
n
m
u
l
t
i
-
c
l
as
s
c
l
a
s
s
i
fi
ca
t
i
o
n
w
i
t
h
ex
p
l
ai
n
ab
l
e
art
i
fi
c
i
a
l
i
n
t
el
l
i
g
en
ce
(X
A
I)
.
O
u
r
p
r
o
p
o
s
e
d
mo
d
e
l
i
n
co
r
p
o
ra
t
es
co
n
v
o
l
u
t
i
o
n
a
l
n
eu
ra
l
n
et
w
o
r
k
(CN
N
)
f
o
r
o
p
erat
i
o
n
al
fea
t
u
re
e
x
t
rac
t
i
o
n
a
n
d
l
o
n
g
s
h
o
r
t
-
t
er
m
memo
ry
(L
ST
M)
t
o
cap
t
u
r
e
t
i
me
-
b
a
s
ed
d
e
p
en
d
en
c
i
es
.
Pred
i
ct
ed
v
al
u
e
i
s
cl
a
s
s
i
f
i
ed
w
i
t
h
mu
l
t
i
c
l
as
s
cl
a
s
s
e
s
-
v
er
y
b
u
l
l
i
s
h
,
b
u
l
l
i
s
h
,
n
eu
t
ral
,
b
ear
i
s
h
,
v
er
y
b
eari
s
h
s
i
g
n
a
l
s
f
o
r
al
l
N
i
ft
y
5
0
s
t
o
c
k
s
.
T
h
e
mo
d
el
i
n
t
eg
ra
t
es
e
s
s
e
n
t
i
a
l
t
ech
n
i
ca
l
i
n
d
i
c
at
o
rs
t
o
fi
n
d
p
at
t
ern
s
fro
m
b
as
i
c
p
ri
ce
d
at
a.
X
A
I
t
ech
n
i
q
u
e
s
are
al
s
o
u
s
e
d
t
o
fi
n
d
feat
u
re
c
o
n
t
ri
b
u
t
i
o
n
s
t
o
mo
d
el
p
r
ed
i
c
t
i
o
n
.
I
t
i
m
p
ro
v
e
s
t
h
e
c
l
ari
t
y
o
f
t
h
e
mo
d
el
’
s
ad
m
i
n
i
s
t
rat
i
v
e
p
ro
ce
d
u
re
b
y
fi
g
u
r
i
n
g
o
u
t
h
o
w
t
ech
n
i
ca
l
i
n
d
i
ca
t
o
r
s
i
n
fl
u
e
n
ce
s
t
o
ck
es
t
i
mat
e
s
.
T
h
e
o
u
t
co
mes
h
i
g
h
l
i
g
h
t
t
h
e
mo
d
e
l
’s
a
b
i
l
i
t
y
t
o
g
en
era
t
e
act
i
o
n
ab
l
e
t
r
ad
i
n
g
s
i
g
n
al
s
,
rei
n
f
o
r
ced
b
y
p
erfo
rma
n
ce
p
r
o
g
re
s
s
me
t
ri
c
s
,
co
n
t
ri
b
u
t
i
n
g
t
o
m
o
re
w
el
l
-
i
n
fo
rme
d
an
d
p
l
a
n
n
e
d
v
e
n
t
u
re
d
ec
i
s
i
o
n
s
.
T
h
e
p
ro
p
o
s
ed
mo
d
el
re
v
eal
s
g
reat
er
p
erfo
rma
n
ce,
reach
i
n
g
an
av
era
g
e
accu
racy
o
f
9
6
%
,
b
ea
t
i
n
g
L
i
g
h
t
G
BM
at
8
9
%
,
ran
d
o
m
fo
res
t
at
8
5
%
,
an
d
s
u
p
p
o
rt
v
ect
o
r
ma
ch
i
n
e
at
6
0
%
.
K
e
y
w
o
r
d
s
:
C
onvolut
ional
ne
ur
a
l
ne
twor
ks
De
e
p
lea
r
ning
E
x
p
l
a
i
na
b
le
a
r
t
i
f
i
c
i
a
l
i
n
te
l
l
ig
e
nc
e
L
ong
s
hor
t
-
ter
m
memor
y
S
tock
pr
e
diction
T
e
c
hnica
l
indi
c
a
tor
s
Th
i
s
is
an
o
p
en
a
c
ces
s
a
r
t
i
c
l
e
u
n
d
e
r
the
CC
BY
-
SA
l
i
ce
n
s
e.
C
or
r
e
s
pon
din
g
A
u
th
or
:
As
hwin
R
a
iyani
I
ns
ti
tut
e
of
M
a
na
ge
ment,
Ni
r
ma
Unive
r
s
it
y
Ahme
da
ba
d,
Guja
r
a
t,
I
ndia
E
mail:
a
s
hwin.
r
kc
e
t@gm
a
il
.
c
om
1.
I
NT
RODU
C
T
I
ON
P
r
e
dicting
the
s
hor
t
-
ter
m
f
utur
e
va
lue
of
pa
r
ti
c
ul
a
r
s
tock
us
ing
dif
f
e
r
e
nt
s
ys
tema
ti
c
a
ppr
oa
c
he
s
is
c
r
it
ica
l
f
o
r
s
hor
t
ter
m
tr
a
ding
,
whic
h
c
ompr
is
e
s
r
igor
ous
his
tor
ica
l
da
ta
a
na
lys
is
a
nd
pa
tt
e
r
n
f
indi
n
g
ins
ide
da
ta
to
pr
oduc
e
pr
of
i
ts
or
mi
nim
ize
los
s
.
B
r
oke
r
s
c
a
n
buy
a
nd
s
e
ll
s
ha
r
e
s
on
the
s
tock
mar
ke
t
to
make
money.
P
r
e
dicting
the
a
c
c
ur
a
te
pos
it
ion
to
buy
,
s
e
ll
,
or
ho
ld
mi
ght
he
lp
b
r
oke
r
s
ge
ne
r
a
te
money
f
r
o
m
the
m
a
r
ke
t
or
mi
nim
ize
los
s
.
E
xa
c
t
pr
ice
moveme
nts
lea
d
to
bi
g
pr
of
it
s
;
so
many
r
e
s
e
a
r
c
he
r
s
a
r
e
int
e
r
e
s
ted
in
t
his
a
r
e
a
.
T
r
a
dit
ional
methodologi
e
s
,
including
as
li
ne
a
r
r
e
gr
e
s
s
ion,
e
xpone
nti
a
l
a
ve
r
a
ging,
a
utor
e
gr
e
s
s
ive
i
ntegr
a
ted
movi
ng
a
ve
r
a
ge
(
AR
I
M
A
)
,
a
nd
ge
ne
r
a
li
z
e
d
a
uto
r
e
gr
e
s
s
ive
c
ondit
ional
he
ter
os
ke
da
s
ti
c
it
y
(
GA
R
C
H
)
,
ha
ve
be
e
n
a
ppli
e
d
f
or
the
pur
pos
e
of
pr
e
di
c
ti
ng
s
tock
pr
ice
s
f
or
a
c
ons
ider
a
ble
a
mount
of
ti
me.
How
e
v
e
r
,
thes
e
a
ppr
oa
c
he
s
a
r
e
li
mi
ted
to
l
inea
r
pa
tt
e
r
ns
a
nd
a
s
s
ume
that
the
da
ta
us
e
a
no
r
mal
dis
tr
ibu
ti
on.
In
r
e
c
e
nt
ye
a
r
s
,
the
domain
of
f
inanc
ial
f
or
e
c
a
s
ti
ng
ha
s
be
e
n
inf
l
ue
nc
e
d
by
a
dva
nc
e
d
mac
hine
lea
r
ning
a
nd
de
e
p
lea
r
ning
models
.
T
he
s
e
models
we
r
e
de
ve
loped
to
ove
r
c
om
e
the
li
mi
tations
of
t
r
a
dit
ional
li
ne
a
r
tec
hniques
,
le
a
ding
to
s
igni
f
ica
nt
im
pr
ove
ments
in
f
or
e
c
a
s
ti
ng
pr
oc
e
s
s
e
s
.
B
oth
types
of
models
a
r
e
c
a
pa
ble
of
ha
nd
li
ng
the
non
-
li
ne
a
r
na
tur
e
of
f
inanc
ial
da
ta
[
1]
,
[
2
]
.
So
he
r
e
in
our
re
s
e
a
r
c
h
we
t
r
ied
to
us
e
dif
f
e
r
e
nt
de
e
p
lea
r
ning
a
nd
mac
hine
lea
r
ning
methods
f
o
r
s
tock
mar
ke
t
pr
e
dict
ion.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
M
ult
i
-
c
las
s
s
tock
mar
k
e
t
for
e
c
as
ti
ng
w
it
h
de
e
p
lear
ning
mode
ls
:
an
e
x
plai
nable
…
(
C
hhay
a
P
atel
)
4343
B
a
s
ic
inf
or
mation
of
s
tock
is
da
y
ope
n
p
r
ice
,
da
y
c
los
e
pr
ice
,
da
y
high
p
r
ice
,
da
y
low
pr
ice
,
a
nd
volum
e
.
Di
f
f
e
r
e
nt
types
of
tec
hnica
l
indi
c
a
tor
s
a
r
e
us
e
d
to
f
ind
unde
r
lyi
ng
pa
tt
e
r
ns
of
s
tock
m
ove
ment.
B
a
s
e
d
on
their
r
oles
,
thes
e
indi
c
a
tor
s
may
be
divi
de
d
int
o
numer
ous
c
a
tegor
ies
:
volum
e
indi
c
a
tor
s
,
mom
e
ntum
indi
c
a
tor
s
,
volatil
it
y
indi
c
a
tor
s
,
a
nd
t
r
e
nd
-
f
oll
owing
indi
c
a
tor
s
.
T
he
mos
t
p
r
omi
ne
nt
t
e
c
hnica
l
indi
c
a
tor
s
include
movi
ng
a
ve
r
a
ge
c
onve
r
ge
nc
e
diver
ge
nc
e
(
M
AC
D)
-
tr
e
nd
-
f
oll
owing
indi
c
a
tor
,
r
e
lative
s
tr
e
ngth
index
(
R
S
I
)
-
mom
e
ntum
indi
c
a
tor
,
a
ve
r
a
ge
tr
ue
r
a
nge
(
AT
R
)
-
volatil
it
y
indi
c
a
tor
,
volum
e
-
we
ight
e
d
a
ve
r
a
ge
p
r
ice
(
VW
AP)
-
c
umul
a
ti
ve
indi
c
a
tor
a
nd
r
a
te
of
c
ha
nge
(
R
OC
)
-
mom
e
ntum
os
c
il
lator
[
3]
,
[
4
]
.
L
o
ng
s
ho
r
t
-
t
e
r
m
me
mo
r
y
(
L
S
T
M
)
o
f
f
e
r
s
a
me
a
s
u
r
a
bl
e
a
p
p
r
o
a
c
h
t
o
s
t
oc
k
ma
r
ke
t
f
o
r
e
c
a
s
t
in
g
a
s
it
is
w
i
de
l
y
us
e
d
f
o
r
ti
me
s
e
r
i
e
s
f
o
r
e
c
a
s
ti
ng
l
ike
s
t
oc
k
da
ta
.
L
S
T
M
mo
de
l
,
a
lo
ng
w
it
h
o
t
he
r
n
e
u
r
a
l
ne
tw
or
k
m
o
de
ls
,
p
e
r
s
is
t
a
n
o
ti
c
e
a
bl
e
f
oc
us
o
f
r
e
s
e
a
r
c
h
a
nd
de
v
e
l
op
me
nt
in
ti
me
s
e
r
ies
f
or
e
c
a
s
t
in
g
.
I
mp
r
ov
e
me
n
ts
i
n
L
S
T
M
w
it
h
o
p
ti
mi
z
a
ti
on
pa
r
a
me
te
r
s
ha
v
e
t
he
po
te
nt
ia
l
t
o
r
a
is
e
s
to
c
k
ma
r
k
e
t
p
r
e
d
ic
t
io
n
c
o
r
r
e
c
t
ne
s
s
.
T
his
is
vi
ta
l
i
n
s
to
c
k
m
a
r
k
e
t
f
o
r
e
c
a
s
t
,
a
s
it
a
l
l
ows
s
to
c
k
ho
ld
e
r
s
a
nd
t
r
a
d
e
r
s
to
me
a
s
u
r
e
t
he
c
ons
is
te
nc
y
o
f
th
e
m
od
e
l
's
f
o
r
e
c
a
s
ts
,
h
e
l
p
in
g
t
he
m
ma
ke
up
-
to
-
d
a
t
e
de
c
is
io
ns
[
5
]
,
[
6
]
.
C
on
vo
lu
ti
on
a
l
ne
ur
a
l
ne
tw
or
ks
(
C
N
Ns
)
wo
r
k
b
y
e
x
te
nd
i
ng
o
ve
r
c
o
nv
ol
ut
io
na
l
f
il
te
r
s
to
in
pu
t
da
ta
,
pe
r
c
e
i
vin
g
na
ti
ve
c
on
f
ig
u
r
a
ti
ons
a
n
d
a
s
s
oc
ia
ti
ons
.
E
a
c
h
l
a
y
e
r
o
f
c
on
vo
lu
t
io
na
l
a
bs
tr
a
c
ts
e
x
pr
e
s
s
i
ve
s
t
r
uc
tu
r
e
s
s
u
c
h
a
s
p
r
ic
e
po
in
ts
,
mo
ve
me
nts
,
o
r
r
e
lat
i
ons
hi
ps
b
e
t
we
e
n
t
e
c
hn
ica
l
in
di
c
a
to
r
s
b
y
s
c
a
n
ni
ng
f
il
te
r
s
a
c
r
os
s
th
e
t
i
me
s
e
r
ies
da
ta
.
T
h
e
p
oo
li
ng
la
ye
r
s
the
n
r
e
d
uc
e
t
he
d
i
me
ns
i
ons
,
p
r
o
te
c
t
in
g
th
e
mos
t
im
po
r
ta
nt
f
e
a
tu
r
e
s
w
h
il
e
i
nh
i
bi
ti
ng
o
ve
r
f
i
t
ti
ng
d
ue
to
t
he
c
u
r
s
e
of
d
i
me
ns
i
ona
l
it
y
[
7
]
.
T
his
r
e
s
e
a
r
c
h
is
mo
t
iva
te
d
by
the
im
pr
ov
e
m
e
n
t
o
f
s
t
oc
k
mo
ve
m
e
n
t
p
r
e
di
c
t
io
n
by
u
s
i
n
g
bo
th
p
r
ic
e
a
n
d
vo
lu
me
b
a
s
e
d
tec
hn
ic
a
l
i
nd
ic
a
t
o
r
s
th
a
t
r
e
pl
ic
a
t
e
ma
r
k
e
t
mo
me
nt
um
a
nd
d
i
r
e
c
t
io
n
.
I
n
th
is
s
t
u
dy
,
we
r
e
c
o
mm
e
nd
a
h
yb
r
id
mo
de
l
t
ha
t
c
om
bi
ne
s
C
NN
s
f
or
f
e
a
tu
r
e
e
x
t
r
a
c
t
io
n
a
n
d
L
S
T
M
f
o
r
c
ons
e
c
u
t
ive
le
a
r
ni
n
g
.
Although
de
e
p
lea
r
ning
models
li
ke
C
NN
s
a
nd
L
S
T
M
s
de
a
l
with
high
pr
e
dictive
a
c
c
ur
a
c
y,
their
blac
k
-
box
tende
nc
y
pos
e
s
an
im
por
tant
e
nc
ounter
in
f
inanc
ial
a
ppli
c
a
ti
ons
.
F
inanc
ial
a
na
lys
ts
ne
e
d
not
only
de
tailed
e
s
ti
mate
s
but
a
ls
o
obvious
c
lar
if
ica
ti
ons
f
or
model
c
onc
lus
ions
.
E
xplaina
ble
a
r
ti
f
icia
l
in
t
e
ll
igenc
e
(
XA
I
)
a
im
s
to
a
s
s
oc
iate
thi
s
ga
p
by
pr
ovidi
ng
vis
ions
int
o
how
models
r
e
a
c
h
their
pr
e
dictions
.
In
th
is
s
tudy,
we
a
ppli
e
d
XAI
methods
:
loca
l
int
e
r
p
r
e
table
m
ode
l
-
a
gnos
ti
c
e
xplana
ti
on
s
(
L
I
M
E
)
a
nd
S
ha
pley
a
ddit
ive
e
xplana
ti
ons
(
S
HA
P
)
.
By
incor
por
a
ti
ng
XAI
tec
h
niques
,
the
s
tudy
im
p
r
ove
s
the
model's
tr
a
ns
pa
r
e
nc
y
a
nd
f
os
ter
s
tr
us
t
a
mong
t
r
a
de
r
s
a
nd
s
take
holder
s
[
8]
–
[
1
0]
.
S
tock
mar
ke
t
pr
e
dicting
ha
s
be
c
ome
a
pr
o
mi
ne
nt
r
e
s
e
a
r
c
h
a
r
e
a
in
the
fi
na
nc
ial
domain
a
nd
is
now
vit
a
l
pa
r
t
of
a
r
ti
f
icia
l
int
e
ll
igenc
e
.
R
e
s
e
a
r
c
he
r
s
a
r
e
us
ing
nume
r
ous
s
tatis
ti
c
a
l
mea
s
ur
e
ments
,
mac
hine
lea
r
ning
a
lgor
it
hms
,
a
nd
va
r
ious
de
e
p
lea
r
ning
a
lg
or
it
hms
to
f
ind
an
unde
r
lyi
ng
pa
tt
e
r
n
f
r
om
s
tock
d
a
ta.
T
he
r
e
late
d
wor
ks
ba
s
e
d
on
the
given
topi
c
s
a
r
e
dis
c
us
s
e
d
he
r
e
with
T
a
ble
1.
T
he
e
xis
ti
ng
wor
ks
no
ted
in
T
a
ble
1
.
P
r
im
a
r
il
y
f
oc
us
on
binar
y
c
las
s
if
ica
ti
on
o
r
ne
xt
-
da
y
c
los
ing
pr
ice
,
whic
h
is
not
s
a
ti
s
f
a
c
tor
y
f
o
r
c
a
ptur
ing
c
ompl
e
x
dyna
mi
c
s
a
nd
pa
r
ti
c
ular
s
of
f
inanc
ial
mar
ke
ts
.
T
r
a
dit
ional
mac
hine
lea
r
ning
a
lgor
it
hms
li
ke
s
uppor
t
ve
c
tor
mac
hine
(
S
VM
)
,
k
-
ne
a
r
e
s
t
ne
ighbor
s
(
kNN
)
,
de
c
is
ion
tr
e
e
(
DT
)
,
r
a
ndom
f
or
e
s
t
(
R
F
)
,
a
nd
A
da
B
oos
t
a
r
e
not
we
ll
ge
ne
r
a
li
z
e
d
with
mul
ti
-
c
las
s
c
las
s
if
ica
ti
on
due
to
a
lac
k
o
f
c
a
pa
bil
it
y
to
ha
nd
le
nonli
ne
a
r
da
tas
e
ts
[
2]
,
[
11]
.
C
ur
r
e
nt
r
e
s
e
a
r
c
h
in
de
e
p
lea
r
ning,
s
uc
h
a
s
L
S
T
M
,
r
e
c
ur
r
e
nt
ne
u
r
a
l
ne
twor
ks
(
R
NN
s
)
,
ha
ve
e
xpos
e
d
notew
or
thy
pr
os
pe
c
ts
in
ti
me
-
s
e
r
ies
pr
e
d
iction
by
e
xc
e
ll
e
ntl
y
c
a
tching
c
hr
onologi
c
a
l
de
pe
nde
nc
ies
withi
n
the
da
ta
[
5]
.
T
a
ble
1.
T
he
c
o
r
r
e
s
ponde
nc
e
of
our
pr
opos
e
d
mod
e
l
with
li
te
r
a
tur
e
R
e
f
e
r
e
nc
e
D
a
ta
s
e
ts
D
ur
a
ti
on
M
e
th
ods
E
va
lu
a
ti
on
me
tr
ic
s
I
nc
or
por
a
ti
on w
it
h X
A
I
[
1]
S
&P
500(
33 c
ompa
ni
e
s
)
2017
-
2021
M
L
P
M
A
E
, R
2
No
[
6]
C
hi
ne
s
e
c
omp
a
ni
e
s
2019
-
2023
L
I
N
E
,
L
S
T
M
M
S
E
, R
M
S
E
,
M
A
P
, R
2
No
[
7]
N
if
ty
50
in
de
x
2014
-
2018
R
N
N
, C
N
N
,
L
S
T
M
M
S
E
, R
M
S
E
, M
A
P
, R
2
No
[
8]
D
A
X
30, F
T
S
E
100
S
&P
500, Nikkie225
1990
-
2022
DNN
A
c
c
ur
a
c
y, F
1
-
s
c
or
e
,
pr
e
c
is
io
n, r
e
c
a
ll
Y
e
s
[
9]
S
&P
500 he
a
lt
h c
a
r
e
pr
ic
e
in
de
x
2017
-
2019
L
S
T
M
M
S
E
, R
M
S
E
Y
e
s
[
11]
N
A
S
D
A
Q
, N
Y
S
E
, F
T
S
E
,
N
I
K
K
E
I
2010
-
2020
R
F
, S
V
M
, kN
N
,
A
N
N
,
A
c
c
ur
a
c
y,
F1
-
s
c
or
e
,
pr
e
c
is
io
n, r
e
c
a
ll
No
[
12]
N
if
ty
50
, S
e
ns
e
x, S
&P
500
2015
-
2021
PSO
-
L
S
T
M
M
S
E
, R
M
S
E
, M
A
P
No
[
13]
C
hi
ne
s
e
s
to
c
k ma
r
ke
t
2018
-
2019
P
C
A
-
L
S
T
M
A
c
c
ur
a
c
y,
F1
-
s
c
or
e
No
[
14]
5 I
ndi
a
n c
ompa
ni
e
s
2007
-
2017
M
L
P
, R
N
N
,
L
S
T
M
, C
N
N
M
A
P
E
No
[
15]
P
in
ga
n B
a
nk
2016
-
2018
P
C
A
-
L
S
T
M
R
M
S
E
, M
A
P
E
No
[
16]
F
in
a
nc
ia
l
ne
w
s
da
ta
2012
-
2016
L
R
, S
V
R
, A
N
N
R
M
S
E
, M
A
E
No
[
17]
N
A
S
D
A
Q
-
100 inde
x
2020
-
2021
PSO
-
ANN
M
S
E
, R
M
S
E
, R
2
No
[
18]
N
if
ty
50
12
-
c
ompa
ni
e
s
2015
-
2021
kN
N
, S
V
M
, D
T
,
L
S
T
M
M
S
E
, R
M
S
E
, M
A
P
, R
2
No
P
r
opos
e
d
mode
l
N
if
ty
50
a
ll
c
ompa
ni
e
s
2020
-
2024
C
N
N
-
L
S
T
M
A
c
c
ur
a
c
y,
F1
-
s
c
or
e
,
pr
e
c
is
io
n, r
e
c
a
ll
Y
e
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14,
No.
5:
Oc
tober
2025
:
43
42
-
4352
4344
Our
p
r
opos
e
d
model
f
oc
us
e
s
on
mul
ti
-
c
las
s
c
las
s
if
ica
ti
on
with
c
las
s
e
s
:
ve
r
y
bull
is
h,
bull
is
h,
ne
utr
a
l
,
be
a
r
is
h,
a
nd
ve
r
y
be
a
r
is
h
s
ignals
f
or
a
ll
Ni
f
ty50
s
tocks
.
As
a
da
tas
e
t,
we
tr
ied
to
f
oc
us
on
a
ll
50
c
ompanie
s
f
r
om
the
Nif
ty50
.
T
o
c
a
ptur
e
a
n
unde
r
lyi
ng
pa
tt
e
r
n
f
r
om
a
given
da
tas
e
t,
we
int
e
gr
a
te
di
f
f
e
r
e
nt
tec
hnica
l
indi
c
a
tor
s
s
uc
h
a
s
R
S
I
,
M
AC
D,
AT
R
,
R
OC
,
VW
AP,
a
nd
many
mor
e
[
6]
.
Our
model
identif
ies
the
be
s
t
c
ombi
na
ti
on
of
tec
hnica
l
indi
c
a
tor
s
a
nd
ge
ne
r
a
tes
5
dif
f
e
r
e
nt
c
las
s
e
s
f
or
p
r
e
diction.
A
C
NN
-
L
S
T
M
f
us
ion
model
de
ve
loped
f
or
thi
s
c
las
s
if
ica
ti
on
[
7]
.
T
o
e
n
ha
nc
e
pr
e
diction
c
a
pa
bil
it
y
a
nd
unde
r
s
tanding
of
models
'
be
ha
vior
,
we
incor
po
r
a
te
the
XA
I
tec
hnique
S
HA
P
a
nd
L
I
M
E
,
whic
h
highl
ight
f
e
a
tur
e
inf
lu
e
nc
e
a
nd
int
e
r
pr
e
tabili
ty.
T
he
s
e
methods
he
lp
us
to
unde
r
s
tand
the
im
por
tanc
e
of
input
f
e
a
tur
e
s
by
pr
ovid
ing
both
global
a
nd
loca
l
int
e
r
p
r
e
tabili
ty.
T
his
leve
l
of
int
e
r
pr
e
tabili
ty
is
c
r
uc
ial
f
or
buil
ding
t
r
us
t
a
mong
s
take
holder
s
a
nd
s
uppor
ti
ng
mor
e
inf
o
r
med
a
nd
c
onf
ident
de
c
is
ion
-
making.
[
8]
,
[
19]
.
M
os
t
e
xis
ti
ng
models
f
oc
us
on
pr
e
diction
a
c
c
ur
a
c
y
but
lac
k
model
tr
a
ns
pa
r
e
nc
y,
whic
h
is
ve
r
y
i
mpo
r
tant
to
unde
r
s
tand
a
nd
tr
us
t
m
ode
l.
Our
r
e
s
e
a
r
c
h
a
ddr
e
s
s
e
s
thi
s
ga
p
by
incor
por
a
ti
ng
C
N
N
-
L
S
T
M
model
with
XA
I
tool
s
,
pr
ovidi
ng
both
a
c
c
ur
a
c
y
a
nd
int
e
r
pr
e
tabili
ty
.
2.
M
E
T
HO
D
In
th
is
r
e
s
e
a
r
c
h
pa
pe
r
,
we
us
e
d
dif
f
e
r
e
nt
tec
hnica
l
indi
c
a
tor
s
to
unc
ove
r
hidden
pa
tt
e
r
ns
ins
ide
ba
s
ic
pr
ice
da
ta.
T
he
a
ddit
ion
of
tec
hnica
l
indi
c
a
tor
s
led
to
a
c
ur
s
e
of
dim
e
ns
ionalit
y,
whic
h
may
ne
ga
ti
ve
ly
inf
luenc
e
model
pe
r
f
or
manc
e
.
To
a
void
thi
s
is
s
u
e
,
C
NN
wa
s
a
ppli
e
d
to
s
olve
the
c
ur
s
e
of
dim
e
n
s
ionalit
y.
Af
ter
f
e
a
tur
e
s
e
lec
ti
on,
the
L
S
T
M
model
wa
s
a
ppli
e
d
c
las
s
if
y
s
tock
mar
ke
t
tr
e
nds
.
We
incor
por
a
ted
XAI
-
S
HA
P
a
nd
L
I
M
E
,
to
highl
ight
f
e
a
tur
e
im
pa
c
t
a
nd
im
pr
ove
int
e
r
pr
e
tabili
ty
.
2.
1
.
Nif
t
y50
c
om
p
an
ies
T
he
Nif
ty50
is
a
major
I
ndian
s
tock
mar
ke
t
inde
x
that
r
e
pr
e
s
e
nts
the
pe
r
f
or
manc
e
of
the
50
mos
t
pr
omi
ne
nt
a
nd
a
c
ti
ve
ly
t
r
a
de
d
c
ompanie
s
on
th
e
na
ti
ona
l
s
tock
e
xc
ha
nge
(
NSE
)
.
It
is
a
s
tanda
r
d
index
e
xtens
ively
us
e
d
by
s
ha
r
e
holder
s
a
nd
r
e
s
e
a
r
c
he
r
s
to
mea
s
ur
e
the
pe
r
f
or
manc
e
of
the
I
nd
ian
e
quit
y
mar
ke
t.
It
include
s
c
or
por
a
ti
ons
f
r
om
dif
f
e
r
e
nt
s
e
c
tor
s
li
ke
f
inanc
e
,
c
ons
umer
goods
,
tec
hnology,
pha
r
mac
e
uti
c
a
ls
,
e
ne
r
gy,
as
s
hown
in
F
igur
e
1
.
C
ompany
s
e
lec
ti
on
is
done
ba
s
e
d
on
mar
ke
t
c
a
pit
a
li
z
a
ti
on
,
l
iqui
dit
y,
tr
a
ding
f
r
e
que
nc
y,
a
nd
index
r
e
ba
lanc
ing
[
13
]
.
F
igur
e
1.
S
e
c
tor
-
wis
e
we
ight
a
ge
in
Nif
ty50
2.
2
.
T
e
c
h
n
ical
in
d
icat
or
s
M
a
r
ke
t
tec
hnica
l
indi
c
a
tor
s
a
r
e
qua
nti
tative
tool
s
t
ha
t
us
e
d
to
identif
y
tr
e
nds
a
s
s
oc
iate
d
with
s
p
e
c
if
ic
s
tocks
.
T
r
a
de
r
s
in
the
ma
r
ke
t
uti
l
ize
thes
e
tr
e
nds
to
f
or
e
c
a
s
t
pr
ice
f
luctua
ti
ons
of
s
tocks
.
I
ndica
tor
s
r
e
ly
on
f
unda
menta
l
s
tock
pr
ice
s
,
including
the
volum
e
of
s
tocks
,
lowe
s
t
p
r
ice
,
ope
ning
pr
ice
,
c
los
ing
p
r
ice
,
a
nd
highes
t
pr
ice
.
T
he
y
of
f
e
r
c
r
it
ica
l
ins
ight
s
a
nd
r
e
ve
a
l
pa
tt
e
r
ns
withi
n
the
da
ta.
By
e
mpl
oying
thes
e
indi
c
a
tor
s
,
br
oke
r
s
unde
r
s
tand
the
ove
r
a
ll
mar
ke
t
s
tr
e
ngth
a
n
d
the
c
ompany’
s
pe
r
f
or
manc
e
.
Our
r
e
s
e
a
r
c
h
ha
s
f
oc
us
e
d
on
c
a
lcula
ti
ng
the
mos
t
inf
luential
tec
hnica
l
indi
c
a
tor
s
,
whic
h
include
the
s
im
ple
movi
ng
a
ve
r
a
ge
(
S
M
A
)
,
R
S
I
,
M
AC
D,
AT
R
,
R
OC
,
a
nd
VW
AP
[
20]
.
T
he
S
M
A
is
c
ons
ider
e
d
by
a
dding
the
c
los
ing
pr
ice
s
of
a
s
toc
k
ove
r
a
li
s
ted
number
of
da
ys
a
nd
then
div
idi
ng
that
tot
a
l
by
the
nu
mber
of
da
ys
as
s
hown
in
(
1)
.
F
or
ins
t
a
nc
e
,
to
de
ter
mi
ne
the
20
-
da
y
S
M
A
of
a
s
tock's
c
los
ing
pr
i
c
e
s
,
one
would
s
um
the
c
los
ing
pr
ice
s
f
r
om
the
las
t
20
da
ys
a
nd
divi
de
the
s
um
by
20
[
5]
.
=
(
1)
T
he
R
S
I
is
a
tec
hnica
l
indi
c
a
tor
wor
king
to
mea
s
ur
e
the
s
pe
e
d
a
nd
pr
ice
a
c
ti
ons
in
s
tocks
.
T
his
indi
c
a
tor
wor
ks
on
the
pr
inciple
that
pr
ice
s
u
s
ua
ll
y
c
ha
nge
ins
ide
a
s
pe
c
if
ic
r
a
nge
unde
r
va
r
yin
g
mar
ke
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
M
ult
i
-
c
las
s
s
tock
mar
k
e
t
for
e
c
as
ti
ng
w
it
h
de
e
p
lear
ning
mode
ls
:
an
e
x
plai
nable
…
(
C
hhay
a
P
atel
)
4345
c
ir
c
ums
tanc
e
s
.
It
he
lps
in
knowing
ove
r
bought
a
nd
ove
r
s
old
s
it
ua
ti
ons
,
in
a
ddit
ion
to
potential
r
e
ve
r
s
a
ls
in
tr
e
nds
.
T
he
c
a
lcula
ti
on
of
the
R
S
I
can
be
pe
r
f
o
r
m
e
d
us
ing
the
RS
va
lue
as
c
a
lcula
ted
in
(
2)
[
3]
.
RS
va
lue
is
ge
ne
r
a
ted
ba
s
e
d
on
the
a
ve
r
a
ge
ga
in
a
nd
a
ve
r
a
ge
l
os
s
of
a
s
pe
c
if
ic
pe
r
iod
as
given
wi
th
(
3)
.
W
it
h
th
e
he
lp
of
(
4)
a
nd
(
5)
,
the
a
ve
r
a
ge
pr
ice
ga
in
a
nd
a
ve
r
a
ge
pr
ice
los
s
ove
r
a
c
e
r
tain
time
pe
r
iod
a
r
e
us
e
d
to
c
om
pute
the
a
ve
r
a
ge
ga
in
a
nd
a
ve
r
a
ge
los
s
.
=
100
−
100
1
+
(
2)
=
(
3)
=
ℎ
ℎ
(
4)
=
ℎ
ℎ
(
5)
P
r
ice
ga
in
de
notes
the
c
ha
nge
be
twe
e
n
the
c
los
i
ng
pr
ice
s
of
the
c
ur
r
e
nt
pe
r
iod
a
nd
the
p
r
e
vious
pe
r
iod.
In
c
ontr
a
s
t,
pr
ice
los
s
s
hows
the
dif
f
e
r
e
nc
e
be
twe
e
n
the
c
los
ing
pr
ice
s
of
the
pr
e
vious
pe
r
iod
a
nd
the
c
ur
r
e
nt
pe
r
iod.
T
he
R
S
I
is
r
e
s
ult
ing
f
r
om
the
a
ve
r
a
ge
of
r
e
c
e
nt
p
r
ice
ga
ins
a
nd
los
s
e
s
ove
r
a
c
hos
e
n
ti
mef
r
a
me.
T
he
us
ua
ll
y
ut
il
ize
d
pe
r
iod
f
or
R
S
I
de
s
ign
is
14
da
ys
[
3
]
.
T
he
M
AC
D
is
a
wide
ly
us
e
d
tec
hnica
l
e
xa
mi
na
ti
on
tool
us
e
d
in
f
inanc
ial
mar
ke
ts
to
c
las
s
if
y
pos
s
ibl
e
c
ha
nge
s
in
mom
e
ntum
,
tr
e
nd
r
e
ve
r
s
a
ls
,
a
nd
s
igns
f
or
buying
or
s
e
ll
ing.
T
he
M
AC
D
is
de
r
i
ve
d
f
r
om
two
e
xpone
nti
a
l
movi
ng
a
ve
r
a
ge
s
(
E
M
As
)
of
s
tock
pr
ice
s
,
e
xa
c
tl
y
the
26
-
da
y
E
M
A
a
nd
the
12
-
d
a
y
E
M
A.
T
he
M
AC
D
li
ne
is
c
a
lcula
t
e
d
by
s
ubtr
a
c
ti
ng
the
26
-
da
y
E
M
A
f
r
om
the
12
-
da
y
E
M
A.
Ana
lys
ts
a
n
d
tr
a
de
r
s
of
ten
e
mpl
oy
the
M
AC
D
togethe
r
with
o
ther
tec
hn
ica
l
indi
c
a
tor
s
to
im
p
r
ove
thei
r
tr
a
ding
de
c
is
ions
[
1
5]
.
T
he
AT
R
is
e
xtens
ively
r
e
c
ognize
d
as
a
mea
s
ur
e
of
volatil
it
y.
It
a
s
s
is
ts
tr
a
de
r
s
in
a
s
s
e
s
s
ing
the
volatil
it
y
of
a
s
tock.
A
higher
AT
R
indi
c
a
tes
that
the
s
tock
is
mor
e
volatil
e
,
while
a
lowe
r
AT
R
s
igni
f
ies
r
e
duc
e
d
volatil
it
y.
T
he
c
a
lcula
ti
on
of
A
T
R
uti
li
z
e
s
the
c
ur
r
e
nt
pe
r
iod's
high,
the
c
ur
r
e
nt
pe
r
iod's
low,
a
nd
the
c
los
ing
pr
ice
f
r
om
the
p
r
e
vious
da
y
as
us
ing
(
6
)
[
3
]
.
=
(
−
1
)
+
(
6)
T
he
B
oll
inger
ba
nds
wor
k
as
a
volatil
it
y
indi
c
a
tor
c
ons
is
ti
ng
of
thr
e
e
li
ne
s
:
the
mi
ddle
ba
nd,
whic
h
is
the
20
-
da
y
S
M
A
s
pe
c
if
ied
as
(
7)
;
the
higher
ba
nd,
c
a
lcula
ted
as
the
mi
ddle
ba
nd
plus
two
ti
mes
the
s
tanda
r
d
de
viation
(
S
D)
s
pe
c
if
ied
as
(
8)
;
a
nd
the
lowe
r
ba
nd,
de
ter
mi
ne
d
by
s
ubtr
a
c
ti
ng
two
ti
me
s
the
SD
f
r
om
the
mi
dd
le
ba
nd
as
s
pe
c
if
ied
(
9)
.
T
he
20
-
da
y
B
oll
inger
ba
nd
is
c
a
lcula
ted
a
c
c
or
dingl
y
[
3]
.
=
(
20
)
(
7)
ℎ
=
+
(
2
×
20
)
(
8)
=
−
(
2
×
20
)
(
9)
2.
3
.
Dim
e
n
s
ion
ali
t
y
r
e
d
u
c
t
ion
wi
t
h
c
on
vol
u
t
ion
al
n
e
u
r
al
n
e
t
wor
k
In
s
tock
mar
ke
t
p
r
e
diction,
we
f
r
e
que
ntl
y
ha
ve
a
lar
ge
number
of
f
e
a
tur
e
s
,
including
ba
s
ic
pr
ice
s
a
nd
dif
f
e
r
e
nt
tec
hnica
l
indi
c
a
tor
s
.
Additi
ona
ll
y
,
we
ha
ve
da
ta
f
r
om
d
if
f
e
r
e
nt
c
ompanie
s
.
High
-
dim
e
ns
ional
da
ta
c
r
e
a
tes
the
c
ur
s
e
of
dim
e
ns
ionalit
y,
tr
igge
r
ing
ove
r
f
it
t
ing
in
the
model.
Dif
f
e
r
e
nt
mac
hine
lea
r
n
i
ng
/
de
e
p
lea
r
ning
models
a
r
e
us
e
d
to
r
e
duc
e
the
c
u
r
s
e
of
dim
e
ns
ionalit
y,
li
ke
p
r
incipa
l
c
omponent
a
na
lys
is
(
P
C
A)
,
a
utoenc
ode
r
s
,
a
nd
C
NN
s
.
C
NN
s
a
r
e
s
uppos
e
d
to
c
ut
thi
s
high
-
dim
e
ns
ional
da
ta
int
o
lowe
r
-
dim
e
ns
ional,
mor
e
de
s
c
r
ipt
ive
da
ta
while
c
ons
e
r
ving
i
mpor
tant
pa
tt
e
r
ns
c
ompar
e
d
with
P
C
A
a
nd
a
utoenc
ode
r
s
.
C
NN
f
inds
non
-
li
ne
a
r
pa
tt
e
r
ns
f
r
om
g
iven
da
ta,
as
c
ompar
e
d
to
P
C
A,
whic
h
wo
r
ks
only
with
li
ne
a
r
da
ta.
C
NN
is
c
ompos
e
d
of
mul
ti
ple
laye
r
s
,
as
il
lus
tr
a
ted
in
F
ig
ur
e
2.
C
onvolut
ional
laye
r
s
a
r
e
us
e
d
f
or
f
e
a
tur
e
e
xtr
a
c
ti
on.
P
ooli
ng
laye
r
s
a
r
e
us
e
d
f
or
d
im
e
ns
ionalit
y
r
e
duc
ti
o
n.
Non
-
li
ne
a
r
it
y
is
int
r
oduc
e
d
us
ing
the
a
c
ti
va
ti
on
f
unc
ti
on
r
e
c
ti
f
ied
li
ne
a
r
unit
(
R
e
L
U)
.
We
can
a
pply
mul
ti
ple
c
onvolut
ional
a
nd
pooli
ng
laye
r
s
to
r
e
c
ognize
pa
tt
e
r
ns
f
r
om
given
da
ta.
T
he
f
inal
output
is
a
ppli
e
d
to
the
f
lattene
d
laye
r
that
c
onve
r
ts
to
a
1D
ve
c
tor
[
16]
,
[
2
1]
.
2.
4
.
T
e
m
p
or
al
m
od
e
li
n
g
wit
h
lon
g
s
h
or
t
t
e
r
m
m
e
m
or
y
L
S
T
M
is
a
s
pe
c
ialize
d
a
r
c
hit
e
c
tur
e
of
R
NN
s
that
e
xc
e
ls
in
de
a
li
ng
with
time
s
e
r
ies
da
ta,
making
it
mainly
us
e
f
ul
in
s
tock
pr
ice
f
or
e
c
a
s
ti
ng.
L
S
T
M
ne
twor
ks
a
r
e
e
nginee
r
e
d
to
e
f
f
icie
ntl
y
c
a
ptur
e
lo
ng
-
ter
m
de
pe
nde
nc
ies
,
de
mons
tr
a
ti
ng
to
be
pa
r
ti
c
ular
ly
he
l
pf
ul
onc
e
e
xa
mi
ning
time
s
e
r
ies
da
ta,
including
p
a
s
t
s
tock
pr
ice
s
.
T
he
de
s
ign
of
an
L
S
T
M
model
s
pe
c
if
ica
ll
y
a
ddr
e
s
s
e
s
the
pr
ob
lems
a
s
s
oc
iate
d
with
lea
r
n
ing
a
nd
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14,
No.
5:
Oc
tober
2025
:
43
42
-
4352
4346
r
e
taining
long
-
ter
m
de
pe
nde
nc
ies
withi
n
s
e
que
nti
a
l
da
tas
e
ts
.
As
s
oc
iate
d
to
c
onve
nti
ona
l
R
NN
s
,
L
S
T
M
s
pos
s
e
s
s
a
mor
e
c
ompl
ica
ted
s
tr
uc
tur
e
,
int
e
r
p
r
e
ti
n
g
them
h
ighl
y
e
f
f
e
c
ti
ve
f
or
a
ppli
c
a
ti
ons
that
r
e
q
uir
e
the
a
c
c
e
pti
ng
of
long
-
ter
m
de
pe
nde
nc
ies
.
An
outl
ine
of
the
main
e
leme
nts
that
make
up
an
L
S
T
M
ne
twor
k's
s
tr
uc
tur
e
is
pr
ovided
with
F
igur
e
3
[
9]
,
[
21]
.
T
he
c
e
ll
s
tate
input
(
C
t
0
)
wor
ks
as
the
long
-
ter
m
memor
y
withi
n
the
L
S
T
M
a
r
c
hit
e
c
tur
e
.
It
p
r
oc
e
s
s
e
s
the
whole
s
tr
uc
tur
e
,
pe
r
mi
tt
ing
the
a
ddit
ion
or
e
l
im
ination
of
da
ta
thr
ough
s
pe
c
if
ica
ll
y
de
s
igned
mec
ha
nis
ms
known
as
ga
tes
.
In
c
ont
r
a
s
t,
the
hidden
s
tate
input
(
Ht
0
)
r
e
pr
e
s
e
nts
the
s
hor
t
-
ter
m
memor
y,
f
or
me
d
by
both
the
c
ur
r
e
nt
input
a
nd
the
c
e
ll
s
tate
.
L
S
T
M
s
e
mpl
oy
thr
e
e
ga
tes
to
a
c
c
ompl
is
h
the
dir
e
c
ti
on
of
da
ta
f
lows
:
input
ga
te,
output
ga
te
,
a
nd
f
or
ge
t
ga
te.
T
he
input
ga
te
ha
ndl
e
s
whic
h
da
ta
pa
s
s
e
d
in
the
c
e
ll
s
tate
,
while
the
f
or
ge
t
ga
te
a
ll
ow
to
r
e
jec
t
inf
or
mation
.
T
he
output
ga
te
no
r
m
a
li
z
e
s
the
f
oll
owing
hidden
s
tate
ba
s
e
d
on
the
a
da
pted
c
e
ll
s
t
a
te
[
10]
,
[
21
]
.
2.
5
.
E
xp
lai
n
ab
le
ar
t
if
icial
in
t
e
ll
igence
XAI
is
ve
r
y
im
por
tant
f
or
AI
dr
iven
models
as
it
incr
e
a
s
e
s
tr
a
ns
pa
r
e
nc
y
a
nd
int
e
r
pr
e
tabili
ty.
It
a
ll
ows
us
to
e
s
tablis
h
c
onf
idenc
e
a
nd
tr
us
t
in
c
o
mpl
e
x
blac
k
-
box
AI
models
.
Give
n
that
s
tock
pr
e
diction
models
c
omm
only
e
xploi
t
int
r
ica
te
mac
hine
lea
r
ning
a
nd
de
e
p
lea
r
ning
a
lgor
it
h
ms
,
it
is
s
igni
f
i
c
a
nt
f
or
tr
a
de
r
s
,
inves
tor
s
,
a
nd
f
inanc
ial
a
na
lys
ts
to
unde
r
s
tand
the
de
c
is
ion
-
making
pr
oc
e
s
s
e
s
be
hind
thes
e
models
.
Dif
f
e
r
e
nt
XAI
methods
,
s
uc
h
as
S
HA
P
,
L
I
M
E
,
a
n
d
g
r
a
dient
-
we
ight
e
d
c
las
s
a
c
ti
va
ti
on
mapping
(
Gr
a
d
-
C
A
M
)
,
a
r
e
us
e
d
to
identif
y
im
por
tant
f
e
a
tur
e
s
in
AI
model
s
[
22]
,
[
23]
.
F
igur
e
2.
Dif
f
e
r
e
nt
laye
r
s
of
C
NN
F
igur
e
3.
L
S
T
M
ne
twor
k's
s
tr
uc
tu
r
e
3.
P
ROP
OS
E
D
M
ODE
L
A
s
umm
a
r
y
of
the
pr
opos
e
d
f
us
ion
model,
incor
p
or
a
ti
ng
L
S
T
M
with
C
NN
,
is
pr
e
s
e
nted
he
r
e
a
long
with
F
igur
e
4.
T
he
pr
opos
e
d
model
take
s
5
ye
a
r
s
of
ope
n,
h
igh,
low,
a
nd
c
los
e
(
OH
L
C
)
r
a
w
da
ta
f
r
om
a
ll
Nif
ty50
c
ompanie
s
as
input
.
F
e
a
tur
e
s
e
lec
ti
on
tec
hniques
a
r
e
a
ppli
e
d
to
e
xtr
a
c
t
ke
y
indi
c
a
tor
s
r
e
l
e
va
nt
to
mar
ke
t
moveme
nts
.
Af
te
r
da
ta
c
lea
ning,
ta
r
ge
t
s
ignals
a
r
e
ge
ne
r
a
ted
to
t
r
a
in
model.
T
he
s
e
lec
ted
f
e
a
tur
e
s
a
r
e
then
f
e
d
int
o
the
L
S
T
M
-
C
NN
f
us
ion
model
f
or
e
f
f
e
c
ti
ve
mar
ke
t
pr
e
diction
.
3.
1
.
Dat
a
c
oll
e
c
t
ion
T
he
da
tas
e
t
f
or
our
r
e
s
e
a
r
c
h
wor
k
is
the
Ni
f
ty50
a
ll
c
ompanie
s
’
his
tor
ica
l
in
f
or
mation
,
including
a
ll
ba
s
ic
OH
L
C
pr
ice
s
a
nd
volum
e
.
T
he
pe
r
iod
of
his
tor
ica
l
da
ta
is
5
ye
a
r
s
,
f
r
o
m
J
a
nua
r
y
1,
2020
to
D
e
c
e
mber
31,
2024
,
so
f
o
r
one
c
ompany,
s
tock
da
ta
r
ows
a
r
e
a
bout
1
,
206;
he
nc
e
,
f
or
a
tot
a
l
of
50
c
ompanie
s
,
c
oll
e
c
ted
r
ows
a
r
e
a
bout
57
,
105.
All
c
ompanie
s
’
da
ta
a
r
e
c
ol
lec
ted
us
ing
the
yf
inanc
e
AP
I
[
2
]
,
[
17]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
M
ult
i
-
c
las
s
s
tock
mar
k
e
t
for
e
c
as
ti
ng
w
it
h
de
e
p
lear
ning
mode
ls
:
an
e
x
plai
nable
…
(
C
hhay
a
P
atel
)
4347
3.
2
.
F
e
a
t
u
r
e
e
n
gin
e
e
r
in
g
To
pr
e
dict
f
utu
r
e
p
r
ice
a
c
ti
ons
,
a
na
lys
ts
ope
r
a
te
tec
hnica
l
indi
c
a
tor
s
,
whic
h
a
r
e
methods
f
or
e
xa
mi
ning
pr
e
vious
pr
ice
a
nd
volum
e
da
ta
of
a
c
ompany.
T
he
s
e
indi
c
a
tor
s
,
whic
h
a
r
e
de
r
i
ve
d
f
r
om
mathe
matica
l
c
a
lcula
ti
ons
,
can
a
s
s
is
t
tr
a
de
r
s
in
making
knowle
dge
a
ble
de
c
is
ions
c
onc
e
r
ning
the
buying/
s
e
ll
ing
of
pa
r
ti
c
ular
s
tocks
.
TA
-
L
ib
li
br
a
r
y
is
us
e
d
to
c
a
lcula
te
tec
hnica
l
indi
c
a
tor
s
f
or
s
to
c
ks
.
Our
r
e
s
e
a
r
c
h
pr
opos
e
d
the
c
a
lcula
ti
on
of
the
mos
t
a
f
f
e
c
ted
tec
hnica
l
indi
c
a
tor
s
,
whic
h
a
r
e
R
S
I
,
M
AC
D,
S
M
As
,
AT
R
,
R
OC
,
a
nd
VW
AP
[
11]
.
F
igur
e
4.
P
r
opos
e
d
C
NN
-
L
S
T
M
model
3.
3
.
E
xp
lorat
or
y
d
at
a
a
n
alys
is
Af
ter
the
c
a
lcula
ti
on
of
tec
hnica
l
indi
c
a
tor
s
,
we
no
w
ha
d
27
f
e
a
tur
e
s
or
c
olum
ns
to
be
c
ons
ider
e
d
f
o
r
da
ta
e
xplor
a
ti
on.
S
M
A_20
da
ys
will
ge
ne
r
a
te
Na
N
va
lue
f
o
r
the
f
i
r
s
t
20
da
ys
of
a
pa
r
ti
c
ula
r
c
om
pa
ny,
the
s
a
me
wa
y
R
S
I
_14
will
ge
ne
r
a
te
Na
N
f
or
the
f
ir
s
t
14
da
ys
.
B
a
s
e
d
on
their
mathe
matica
l
f
or
mul
a
ti
ons
,
the
a
ppli
e
d
tec
hnica
l
indi
c
a
tor
s
inher
e
ntl
y
ge
ne
r
a
te
Na
N
va
lues
f
or
the
ini
ti
a
l
pe
r
iods
.
To
make
da
ta
r
e
a
dy
f
or
C
NN
,
we
f
il
led
the
Na
N
va
lue
by
us
ing
f
o
r
wa
r
d
a
n
d
then
ba
c
kwa
r
d
methods
.
3.
4
.
Gener
at
e
t
ar
ge
t
c
las
s
e
s
b
as
e
d
on
p
e
r
f
or
m
a
n
c
e
in
d
icat
or
s
As
we
im
pleme
nted
mul
ti
-
c
las
s
c
las
s
if
ica
ti
on,
we
ha
d
to
ge
ne
r
a
te
a
tot
a
l
of
5
c
las
s
e
s
ba
s
e
d
on
pe
r
f
or
manc
e
indi
c
a
tor
s
.
We
c
a
lcula
ted
the
a
ve
r
a
g
e
of
the
pe
r
c
e
ntage
r
e
tu
r
ns
a
c
r
o
ss
1,
5,
10
,
20
,
a
n
d
30
da
ys
f
or
each
r
ow
in
the
da
tas
e
t,
a
nd
we
s
a
ve
d
the
out
c
ome
in
the
"
Ave
r
a
ge
_R
e
tur
n"
c
olum
n
as
de
s
c
r
ib
e
d
(
10)
.
B
a
s
e
d
on
Ave
r
a
ge
_R
e
tur
n,
we
c
las
s
if
y
each
r
ow
in
to
one
of
f
ive
c
las
s
e
s
de
s
c
r
ibed
in
T
a
ble
2.
_
=
1
+
5
+
10
+
20
+
30
5
(
10)
T
a
ble
2
.
Ge
ne
r
a
te
tar
ge
t
c
las
s
e
s
ba
s
e
d
on
a
ve
r
a
ge
_r
e
tur
n
indi
c
a
tor
s
C
ondi
ti
on
T
a
r
ge
t
la
be
l
M
e
a
ni
ng
A
ve
r
a
ge
_R
e
tu
r
n >
5
C
la
s
s
1
V
e
r
y bull
is
h
2 ≤A
ve
r
a
ge
_R
e
tu
r
n ≤5
C
la
s
s
2
B
ul
li
s
h
-
2
<
A
ve
r
a
ge
_R
e
tu
r
n <
2
C
la
s
s
0
N
e
ut
r
a
l
-
5 ≤A
ve
r
a
ge
_R
e
tu
r
n ≤
-
2
C
la
s
s
3
B
e
a
r
is
h
A
ve
r
a
ge
_R
e
tu
r
n <
-
5
C
la
s
s
4
V
e
r
y be
a
r
is
h
3.
5
.
Dat
a
p
r
e
p
r
oc
e
s
s
in
g
As
pa
r
t
of
da
ta
pr
e
pr
oc
e
s
s
ing,
the
da
ta
is
dis
tr
ibut
e
d
int
o
f
e
a
tur
e
s
a
nd
tar
ge
t
va
r
iable
s
.
T
e
c
hnica
l
indi
c
a
tor
s
a
r
e
us
e
d
as
f
e
a
tu
r
e
s
,
while
the
tar
ge
t
c
ons
is
ts
of
f
ive
d
if
f
e
r
e
nt
c
las
s
e
s
r
e
pr
e
s
e
nti
ng
va
r
io
us
s
tock
pr
ice
a
c
ti
vit
ies
.
Da
ta
s
c
a
li
ng
is
vit
a
l
to
s
tanda
r
dize
the
f
e
a
tur
e
s
,
gua
r
a
ntee
ing
they
f
a
ll
ins
ide
a
pr
e
c
is
e
s
e
r
ies
,
whic
h
he
lps
incr
e
a
s
e
the
pe
r
f
or
manc
e
a
nd
c
onve
r
g
e
nc
e
of
de
e
p
lea
r
ning
pr
otot
ype
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14,
No.
5:
Oc
tober
2025
:
43
42
-
4352
4348
3.
6
.
CNN
-
L
S
T
M
m
od
e
l
ar
c
h
it
e
c
t
u
r
e
Our
pr
opos
e
d
model
int
e
gr
a
tes
a
c
ombi
na
ti
on
of
the
C
NN
with
L
S
T
M
a
r
c
hit
e
c
tur
e
to
e
nha
nc
e
the
c
ompete
nc
y
of
lea
r
ning
both
s
pa
ti
a
l
a
nd
tempor
a
l
f
e
a
tur
e
s
f
r
om
ti
me
s
e
r
ies
s
tock
da
ta.
F
ir
s
t,
we
a
ppli
e
d
c
onvolut
ional
laye
r
to
f
ind
an
unde
r
lyi
ng
pa
tt
e
r
n
f
r
om
tec
hnica
l
indi
c
a
tor
s
.
T
he
n
the
maxpooling
laye
r
is
a
ppli
e
d
to
r
e
duc
e
dim
e
ns
ionalit
y
by
a
bs
or
bing
t
he
max
va
lue
f
r
om
each
f
il
ter
’
s
output
.
Af
ter
C
NN,
the
L
S
T
M
model
wa
s
a
ppli
e
d
to
c
a
ptur
e
tempo
r
a
l
de
pe
nde
nc
ies
ove
r
ti
me.
To
e
ns
ur
e
the
model
doe
s
not
s
uf
f
e
r
f
r
om
ove
r
f
it
ti
ng
,
r
e
gular
iza
ti
on
is
us
e
d
dur
ing
the
tr
a
ini
ng
pha
s
e
.
T
he
n
f
lattene
d
laye
r
f
oll
owe
d
de
ns
e
laye
r
a
ppli
e
d.
F
inally,
the
output
laye
r
with
the
S
of
tM
a
x
a
c
ti
va
ti
on
f
unc
ti
on
a
ppli
e
d
f
o
r
mul
ti
c
las
s
c
las
s
if
ica
ti
on.
To
make
c
ompar
is
ons
be
twe
e
n
e
xis
ti
ng
models
,
we
s
e
lec
ted
S
VM
,
R
F
,
a
nd
L
ight
GB
M
as
ba
s
e
models
.
De
tails
of
model
a
r
c
hit
e
c
tu
r
e
a
nd
t
r
a
ini
ng
c
onf
igu
r
a
ti
ons
a
r
e
given
with
T
a
ble
3.
T
a
ble
3
.
C
NN
-
L
S
T
M
model
a
r
c
hit
e
c
tur
e
pa
r
a
mete
r
s
L
a
ye
r
/T
a
s
k
P
a
r
a
me
te
r
s
1D
c
onvolut
io
na
l
T
im
e
s
e
r
ie
s
i
nput
de
r
iv
e
d f
r
om t
e
c
hni
c
a
l
in
di
c
a
to
r
s
M
a
xP
ool
in
g1D
64 f
il
te
r
s
, ke
r
ne
l
s
iz
e
=
3, a
c
ti
va
ti
on=
R
e
L
U
, P
ool
s
iz
e
=
2, us
e
d t
o r
e
duc
e
di
me
ns
io
ns
L
S
T
M
50 unit
s
, r
e
tu
r
n_s
e
que
nc
e
s
=
T
r
ue
, c
a
pt
ur
e
s
s
e
que
nt
ia
l
de
p
e
nde
n
c
ie
s
D
r
opout
D
r
opout r
a
te
=
0.2, us
e
d t
o pr
e
ve
nt
ove
r
f
it
ti
ng
F
la
tt
e
n
F
la
tt
e
ns
3D
out
put
i
nt
o 1D
f
or
t
he
de
ns
e
l
a
ye
r
D
e
ns
e
F
ul
ly
c
onne
c
te
d l
a
ye
r
, a
c
ti
va
ti
on=
R
e
L
U
O
ut
put
la
ye
r
D
e
ns
e
l
a
ye
r
w
it
h
S
of
tM
a
x
a
c
ti
va
ti
on f
or
mul
ti
c
la
s
s
c
la
s
s
if
ic
a
ti
on
C
ompi
la
ti
on
O
pt
im
iz
e
r
:
A
da
m;
L
os
s
:
c
a
te
gor
ic
a
l
c
r
os
s
e
nt
r
opy
T
r
a
in
in
g
c
onf
ig
.
E
poc
hs
:
50;
B
a
tc
h
s
iz
e
:
32;
V
a
li
da
ti
on
s
tr
a
te
gy
W
a
lk
-
f
or
w
a
r
d va
li
da
ti
on
(
24 I
te
r
a
ti
ons
)
T
r
a
in
in
g
w
in
dow
E
xpa
ndi
ng t
im
e
w
in
dow
f
r
om J
a
n 2020 to cur
r
e
nt
t
e
s
t
mont
h
T
e
s
ti
ng
w
in
dow
I
mm
e
di
a
te
ne
xt
mont
h (
e
.g., J
a
n 2023, F
e
b 2023, ..., J
a
n 2025)
T
ot
a
l
it
e
r
a
ti
ons
24 w
a
lk
-
f
or
w
a
r
d s
pl
it
s
pe
r
c
ompa
ny (
1
,
200 tot
a
l
it
e
r
a
ti
ons
(
24 s
pl
it
s
×
50 c
ompa
ni
e
s
)
)
E
va
lu
a
ti
on
me
tr
ic
s
A
c
c
ur
a
c
y,
pr
e
c
is
io
n
,
r
e
c
a
ll
, F
1
-
s
c
or
e
, M
a
c
r
o
-
a
ve
r
a
ge
A
U
C
, R
O
C
c
ur
ve
a
na
ly
s
i
s
3.
7
.
E
xp
lai
n
ab
le
ar
t
if
icial
in
t
e
ll
igence
in
t
e
gr
at
i
on
f
or
i
n
t
e
r
p
r
e
t
ab
il
i
t
y
T
his
r
e
s
e
a
r
c
h
pa
pe
r
us
e
d
XA
I
tec
hniques
f
o
r
s
tock
mar
ke
t
p
r
e
diction.
We
a
ppli
e
d
L
I
M
E
a
nd
S
HA
P
.
It
make
s
model
tr
a
ns
pa
r
e
nt
a
nd
unde
r
s
tanda
bl
e
e
s
pe
c
ially
c
ompl
e
x
model
li
ke
C
NN
-
L
S
T
M
model.
AI
models
pe
r
f
or
manc
e
s
li
ke
blac
k
-
box,
they
give
good
pe
r
f
or
manc
e
but
it
is
unc
lea
r
a
bout
how
they
r
e
a
c
he
d
to
that
de
c
is
ion.
XAI
ope
n
the
blac
k
box
a
nd
he
l
p
tr
a
de
r
s
to
unde
r
s
tand
a
nd
tr
us
t
outpu
t
of
model
.
L
I
M
E
f
oc
us
on
loca
l
e
xplana
ti
on
while
S
HA
P
gives
both
loca
l
a
nd
global
ins
ide
da
ta.
It
gives
c
ontr
ibut
io
n
of
each
tec
hnica
l
indi
c
a
tor
in
pr
e
diction.
By
us
ing
them,
we
b
r
idge
the
ga
p
be
twe
e
n
model
pe
r
f
o
r
m
a
nc
e
a
nd
int
e
r
pr
e
tabili
ty,
gua
r
a
ntee
ing
AI
-
dr
iven
f
inanc
ial
p
r
e
dictions
is
both
a
c
c
ur
a
te
a
nd
tr
a
ns
pa
r
e
nt
[
24]
–
[
2
6]
.
3.
8
.
M
od
e
l
e
val
u
at
ion
As
pr
incipa
l
metr
ics
f
or
c
las
s
if
ica
ti
on,
we
c
a
lcula
ted
c
onf
us
ion
matr
ix
a
nd
f
r
om
it
we
de
r
ived
a
c
c
ur
a
c
y,
pr
e
c
is
ion,
r
e
c
a
ll
,
F1
-
s
c
or
e
,
R
OC
c
ur
v
e
,
a
nd
AUC
va
lue.
Ac
c
ur
a
c
y
de
a
li
ngs
the
p
r
op
or
ti
on
of
c
or
r
e
c
tl
y
c
las
s
if
ied
ins
tanc
e
s
(
both
t
r
ue
pos
it
ive
a
nd
t
r
ue
ne
ga
ti
ve
)
ove
r
the
tot
a
l
number
of
i
ns
tanc
e
s
.
P
r
e
c
is
ion
c
a
lcula
tes
the
pr
opor
ti
on
of
pr
ope
r
ly
p
r
e
dicte
d
pos
it
ive
ins
tanc
e
s
out
of
a
ll
ins
tanc
e
s
pr
e
dicte
d
as
pos
it
ive.
R
e
c
a
ll
r
e
f
lec
ts
the
pr
opor
ti
on
of
a
c
tual
pos
it
ive
ins
tanc
e
s
that
we
r
e
pr
ope
r
ly
r
e
c
ognize
d
by
the
model.
F1
-
s
c
or
e
is
the
ha
r
moni
c
mea
n
of
pr
e
c
is
io
n
a
nd
r
e
c
a
ll
.
Additi
ona
ll
y,
we
pr
e
s
e
nt
c
onf
us
ion
matr
ice
s
a
nd
R
OC
c
ur
ve
s
to
f
ur
the
r
e
va
luate
the
model's
pr
e
dictive
c
a
pa
bil
it
ies
.
F
or
f
or
mu
la
r
e
f
e
r
T
a
ble
4.
T
a
ble
4.
M
ode
l
e
va
luation
metr
ics
M
e
tr
ic
s
F
or
mul
a
A
c
c
ur
a
c
y
+
+
+
+
P
r
e
c
is
io
n
+
R
e
c
a
ll
+
F1
-
s
c
or
e
2
×
−
+
4.
RE
S
UL
T
AN
AL
YS
I
S
We
we
ighed
pr
opos
e
d
model
a
nd
be
nc
hmar
k
mod
e
ls
pe
r
f
or
manc
e
us
ing
a
c
c
ur
a
c
y,
pr
e
c
is
ion,
r
e
c
a
ll
,
F1
-
s
c
or
e
,
AUC
,
a
nd
R
OC
c
ur
ve
.
We
e
xe
c
uted
wa
lk
-
f
or
wa
r
d
v
a
li
da
ti
on
(
24
it
e
r
a
ti
ons
)
outc
omes
to
s
tatis
ti
c
a
ll
y
r
e
late
the
pr
opos
e
d
model
with
ba
s
e
models
.
Additi
ona
ll
y,
d
is
ti
nguis
hing
the
r
is
ing
a
tt
e
nti
on
in
model
int
e
r
pr
e
tabili
ty
,
we
us
e
d
L
I
M
E
methods
to
e
xa
mi
ne
the
model’
s
de
c
is
ion
-
making
f
ounda
ti
on
as
we
ll
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
M
ult
i
-
c
las
s
s
tock
mar
k
e
t
for
e
c
as
ti
ng
w
it
h
de
e
p
lear
ning
mode
ls
:
an
e
x
plai
nable
…
(
C
hhay
a
P
atel
)
4349
4.
1
.
P
e
r
f
or
m
an
c
e
e
valu
at
io
n
of
CNN
-
L
S
T
M
m
od
e
l
T
he
pe
r
f
o
r
manc
e
of
the
pr
opos
e
d
model
wa
s
e
va
luate
d
a
ga
ins
t
be
nc
hmar
k
models
(
S
VM
,
R
F
,
a
nd
L
ight
GB
M
)
us
ing
c
las
s
if
ica
ti
on
metr
ics
a
c
c
ur
a
c
y,
pr
e
c
is
ion,
r
e
c
a
ll
,
F1
-
s
c
or
e
,
a
nd
mac
r
o
-
a
ve
r
a
ge
AU
C
,
a
ve
r
a
ge
d
a
c
r
os
s
the
W
F
V
s
tr
a
tegie
s
.
T
he
RF
mod
e
l
a
nd
L
ight
GB
M
a
c
hieve
s
outs
t
a
nding
pe
r
f
or
man
c
e
due
to
its
c
oll
a
bor
a
ti
ve
a
ppr
oa
c
h
,
while
S
VM
whic
h
h
a
ve
s
im
pler
s
tr
uc
tur
e
s
,
s
how
les
s
im
pr
e
s
s
ive
pr
e
dictive
a
bil
it
ies
.
Our
a
nti
c
ipate
d
model
be
a
ts
e
xis
ti
ng
mod
e
ls
,
a
c
c
ompl
is
hing
the
highes
t
a
c
c
ur
a
c
y.
A
c
ompr
e
he
ns
ive
c
ompar
is
on
of
the
models
is
pr
e
s
e
nted
in
T
a
ble
5.
T
a
ble
5
.
C
ompar
is
on
of
a
c
c
ur
a
c
y,
pr
e
c
is
ion,
r
e
c
a
ll
,
F1
-
s
c
or
e
a
nd
AUC
M
ode
l
A
c
c
ur
a
c
y
(%)
P
r
e
c
is
io
n
(%)
R
e
c
a
ll
(%)
F1
-
s
c
or
e
(%)
M
a
c
r
o
-
a
ve
r
a
ge
AUC
(%)
S
V
M
65
62
64
66
51
RF
89
95
92
89
91
L
ig
ht
G
B
M
89
94
91
88
92
P
r
opos
e
d
mode
l
97
96
96
97
96
4.
2
.
S
t
at
is
t
ical
s
ign
i
f
icance
t
e
s
t
in
g
To
make
s
ur
e
the
r
e
li
a
bil
it
y
of
the
pe
r
f
or
man
c
e
im
pr
ove
ments
,
we
c
a
r
r
ied
out
pa
i
r
e
d
t
-
tes
ts
c
ompar
ing
the
pr
opos
e
d
C
NN
-
L
S
T
M
model
w
it
h
ba
s
e
models
a
c
r
os
s
the
outcome
s
of
wa
lk
f
or
wa
r
d
va
li
da
ti
on.
As
given
in
T
a
ble
6,
a
ll
c
ompar
is
on
s
pr
oduc
e
d
p
-
va
lues
unde
r
0.
05
.
T
his
c
onf
i
r
ms
that
the
im
pr
ove
ments
of
the
pr
opos
e
d
model
a
r
e
s
tatis
ti
c
a
ll
y
s
igni
f
ica
nt
a
nd
not
due
to
r
a
ndom
c
ha
nc
e
.
T
a
ble
6.
S
tatis
ti
c
a
l
s
igni
f
ica
nc
e
tes
ti
ng
C
ompa
r
is
on
M
e
tr
ic
t
-
S
ta
ti
s
ti
c
p
-
V
a
lu
e
S
ig
ni
f
ic
a
nc
e
(
(
p <
0.05)
?
)
P
r
opos
e
d
mode
l
vs
.
S
V
M
A
c
c
ur
a
c
y
86.06
0.00000
Y
e
s
P
r
e
c
is
io
n
51.01
0.00000
Y
e
s
R
e
c
a
ll
64.91
0.00000
Y
e
s
F1
-
s
c
or
e
82.32
0.00000
Y
e
s
P
r
opos
e
d
mode
l
vs
.
RF
A
c
c
ur
a
c
y
20.85
0.00003
Y
e
s
P
r
e
c
is
io
n
4.00
0.01613
Y
e
s
R
e
c
a
ll
21.00
0.00003
Y
e
s
F1
-
s
c
or
e
21.92
0.00003
Y
e
s
P
r
opos
e
d
mode
l
vs
.
L
ig
ht
G
B
M
A
c
c
ur
a
c
y
25.30
0.00001
Y
e
s
P
r
e
c
is
io
n
3.14
0.03492
Y
e
s
R
e
c
a
ll
∞
0.00000
Y
e
s
F1
-
s
c
or
e
28.46
0.00001
Y
e
s
t=
∞
me
a
n
s
no va
r
ia
nc
e
i
n t
he
di
f
f
e
r
e
nc
e
va
lu
e
s
4.
3
.
CNN
-
L
S
T
M
ROC
c
u
r
ve
T
he
R
OC
c
ur
ve
is
il
lus
tr
a
ti
on
of
the
t
r
ue
pos
it
ive
r
a
te
a
nd
the
f
a
ls
e
pos
it
ive.
T
he
R
OC
c
ur
ve
he
lps
to
pictur
e
how
we
ll
a
c
las
s
if
ier
is
pe
r
f
or
mi
ng
a
c
r
os
s
dif
f
e
r
e
nt
c
las
s
if
ica
ti
on
thr
e
s
holds
.
F
igur
e
5
s
ho
ws
R
OC
c
ur
ve
s
f
or
p
r
opos
e
d
mul
ti
-
c
las
s
model.
F
igur
e
5.
R
OC
c
ur
ve
s
f
or
pr
opos
e
d
model
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14,
No.
5:
Oc
tober
2025
:
43
42
-
4352
4350
4.
4
.
XAI
-
f
e
at
u
r
e
c
on
t
r
ib
u
t
ion
u
s
in
g
L
I
M
E
We
us
e
d
L
I
M
E
to
unde
r
s
tand
the
pr
e
dictions
of
our
C
NN
-
L
S
T
M
model
f
or
s
tock
ma
r
ke
t
a
na
lys
is
us
ing
the
Ni
f
ty50
da
tas
e
t.
L
I
M
E
of
f
e
r
s
loca
l
int
e
r
pr
e
tabili
ty,
r
e
ve
a
li
ng
how
ke
y
f
e
a
tu
r
e
s
inf
luenc
e
i
ndivi
dua
l
pr
e
dictions
.
As
s
hown
in
F
igur
e
6,
it
s
ugge
s
ts
tr
a
de
r
s
a
s
tr
ong
obs
e
r
va
ti
on
of
the
model's
de
c
is
ion
-
making
pr
oc
e
s
s
.
T
he
outcome
s
s
how
that
20_Da
ys
(
%
)
,
S
M
A_5,
30_Da
ys
(
%
)
,
a
nd
S
M
A_10
play
the
maximum
im
por
tant
r
oles
in
de
te
r
mi
ning
the
model's
r
e
s
ol
uti
ons
.
T
h
is
pr
opos
e
s
that
the
model
s
e
r
ious
ly
t
r
us
ts
on
s
hor
t
-
ter
m
pr
ice
volatil
it
y
whe
n
making
p
r
e
diction
s
.
F
igur
e
6.
F
e
a
tur
e
c
ontr
ibut
ion
us
ing
L
I
M
E
4.
5
.
B
ac
k
t
e
s
t
in
g
of
m
od
e
l
wi
t
h
u
n
s
e
e
n
d
at
a
To
e
va
luate
the
r
e
a
l
-
wor
ld
us
e
f
ulnes
s
of
the
pr
opos
e
d
model,
ba
c
ktes
ti
ng
wa
s
pe
r
f
or
med
us
ing
uns
e
e
n
da
ta
f
r
om
J
a
nua
r
y
1
to
31
,
2025.
We
c
ompar
e
d
pr
e
dictions
-
r
a
nging
f
r
om
ve
r
y
bull
is
h
to
ve
r
y
be
a
r
is
h
-
with
a
c
tual
outcome
s
.
F
inanc
ial
indi
c
a
tor
s
s
uc
h
as
r
e
tur
n
on
inves
tm
e
nt
(
R
OI
)
,
mea
s
ur
e
d
be
twe
e
n
10%
a
nd
15%
,
s
howe
d
that
the
model
wa
s
pr
of
i
table
a
nd
r
obus
t
e
nough
to
be
us
e
d
f
o
r
s
hor
t
-
ter
m
s
tock
pr
e
dictions
in
r
e
a
l
-
wor
ld
t
r
a
ding
s
it
ua
ti
ons
.
5.
CONC
L
USI
ON
In
thi
s
r
e
s
e
a
r
c
h
wor
k,
we
im
pleme
nted
C
NN
-
L
S
T
M
model
f
or
s
tock
mar
ke
t
pr
e
diction
,
by
us
ing
tec
hnica
l
indi
c
a
tor
s
s
uc
h
as
S
M
A,
BB,
A
T
R
,
R
S
I
,
M
AC
D,
a
nd
VW
AP.
T
he
model
a
c
hieve
d
an
a
c
c
ur
a
c
y
of
96%
,
de
mons
tr
a
ti
ng
its
e
f
f
e
c
ti
ve
ne
s
s
in
c
a
tching
mul
ti
laye
r
e
d
pa
tt
e
r
ns
a
nd
moveme
nts
in
s
to
c
k
pr
ice
pr
e
diction,
outper
f
o
r
mi
ng
tr
a
dit
ional
methods
.
To
im
p
r
ove
tr
a
ns
pa
r
e
nc
y,
we
a
ppli
e
d
L
I
M
E
f
or
loca
l
int
e
r
pr
e
tabili
ty,
r
e
ve
a
li
ng
that
s
hor
t
-
ter
m
vo
latil
it
y
indi
c
a
tor
s
li
ke
20
-
Da
ys
%
,
S
M
A_5,
10
-
Da
ys
%
,
S
M
A_10,
R
S
I
14,
playe
d
the
mos
t
s
igni
f
ica
nt
r
oles
in
the
mo
de
l's
pr
e
dictions
.
T
his
wor
k
highl
ight
s
the
potentia
l
of
de
e
p
lea
r
ning
a
nd
X
AI
in
f
inanc
ial
de
c
is
ion
-
making
.
By
including
tec
hnica
l
indi
c
a
tor
s
a
nd
int
e
r
pr
e
tabili
ty
methods
,
our
methodology
e
nha
nc
e
s
tr
a
ns
pa
r
e
nc
y
in
s
tock
mar
ke
t
pr
e
dictions
.
Our
model
wil
l
he
lp
tr
a
de
r
to
mana
ge
r
is
k
a
nd
maximi
z
e
po
r
tf
oli
o
.
6.
F
UT
UR
E
RE
COM
M
E
ND
AT
I
ONS
F
o
r
f
u
tu
r
e
w
o
r
k
,
we
a
im
to
i
nc
or
po
r
a
t
e
mac
r
oe
c
on
o
mi
c
f
e
a
t
u
r
e
s
a
n
d
r
e
a
l
-
ti
me
d
a
t
a
to
f
u
r
th
e
r
bo
os
t
a
na
ly
ti
c
a
l
pe
r
f
o
r
man
c
e
a
n
d
a
p
pl
ic
a
b
il
i
ty
i
n
dy
na
mi
c
s
toc
k
ma
r
ke
ts
.
W
o
r
k
in
g
wi
t
h
r
e
a
l
-
ti
me
d
a
ta
w
i
ll
a
l
l
ow
o
u
r
m
o
de
l
to
lea
r
n
s
ud
de
n
c
h
a
n
ge
i
n
ma
r
ke
t
,
c
u
lt
iv
a
t
in
g
d
e
c
is
i
on
-
mak
in
g
f
o
r
t
r
a
de
r
s
.
F
u
r
t
he
r
mo
r
e
,
me
r
g
in
g
l
oc
a
l
a
s
w
e
l
l
a
s
gl
ob
a
l
ne
w
s
da
ta
s
e
n
ti
me
nts
pe
r
ha
ps
wi
l
l
i
mp
r
ov
e
t
he
a
pp
r
op
r
ia
te
a
c
c
e
pt
in
g
o
f
ma
r
k
e
t
be
ha
vi
or
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
M
ult
i
-
c
las
s
s
tock
mar
k
e
t
for
e
c
as
ti
ng
w
it
h
de
e
p
lear
ning
mode
ls
:
an
e
x
plai
nable
…
(
C
hhay
a
P
atel
)
4351
F
UN
DI
NG
I
NF
ORM
AT
I
ON
Author
s
s
tate
ther
e
is
no
f
unding
invol
ve
d.
AU
T
HO
R
CONT
RI
B
U
T
I
ONS
S
T
AT
E
M
E
N
T
T
his
jour
na
l
us
e
s
the
C
ontr
ibut
o
r
R
oles
T
a
xo
nomy
(
C
R
e
diT
)
to
r
e
c
ognize
indi
vidual
a
uthor
c
ontr
ibut
ions
,
r
e
duc
e
a
utho
r
s
hip
dis
putes
,
a
nd
f
a
c
il
it
a
te
c
oll
a
bor
a
ti
on.
Nam
e
of
Au
t
h
or
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
C
hha
ya
P
a
tel
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
As
hwin
R
a
iyani
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
C
:
C
onc
e
pt
ua
li
z
a
ti
on
M
:
M
e
th
odol
ogy
So
:
So
f
twa
r
e
Va
:
Va
li
da
ti
on
Fo
:
Fo
r
ma
l
a
na
ly
s
is
I
:
I
nve
s
ti
ga
ti
on
R
:
R
e
s
our
c
e
s
D
:
D
a
ta
C
ur
a
ti
on
O
:
W
r
it
in
g
-
O
r
ig
in
a
l
D
r
a
f
t
E
:
W
r
it
in
g
-
R
e
vi
e
w
&
E
di
ti
ng
Vi
:
Vi
s
ua
li
z
a
ti
on
Su
:
Su
pe
r
vi
s
io
n
P
:
P
r
oj
e
c
t
a
dmi
ni
s
tr
a
ti
on
Fu
:
Fu
ndi
ng a
c
qui
s
it
io
n
CONF
L
I
CT
OF
I
NT
E
RE
S
T
S
T
AT
E
M
E
N
T
Author
s
s
tate
no
c
onf
li
c
t
of
int
e
r
e
s
t.
I
NF
ORM
E
D
CONSE
NT
T
his
s
tudy
did
not
invol
ve
indi
viduals
or
a
ny
pe
r
s
ona
l
identif
ica
ti
on
inf
or
mation
that
c
ould
r
e
quir
e
a
ny
inf
or
med
c
ons
e
nt.
E
T
HI
CA
L
AP
P
ROVA
L
T
his
pa
pe
r
doe
s
not
invol
ve
pe
ople
o
r
a
nim
a
ls
;
no
inves
ti
ga
ti
on
ha
s
invol
ve
d
human
s
ubjec
ts
.
T
he
r
e
f
or
e
,
the
a
utho
r
s
did
not
s
e
e
k
a
ppr
ova
l
f
r
om
a
ny
ins
ti
tut
ional
r
e
view
b
oa
r
d.
DA
T
A
AV
AI
L
A
B
I
L
I
T
Y
T
his
s
tudy
uti
li
z
e
d
his
tor
ica
l
s
tock
pr
ice
a
nd
volum
e
da
ta
f
or
a
ll
50
c
ompanie
s
include
d
in
the
Nif
ty50
index
.
T
he
da
tas
e
t
is
publi
c
ly
a
c
c
e
s
s
ibl
e
thr
ough
the
Na
ti
ona
l
S
tock
E
xc
ha
nge
o
f
I
ndia
(
N
S
E
)
a
nd
ope
n
f
inanc
ial
da
ta
platf
o
r
ms
s
uc
h
a
s
Ya
hoo
F
ina
nc
e
.
R
e
s
e
a
r
c
he
r
s
c
a
n
obtain
the
da
ta
dir
e
c
tl
y
f
r
om
the
NSE
of
f
icia
l
we
bs
it
e
a
t
ht
tps
:/
/www
.
ns
e
indi
a
.
c
om/
o
r
f
r
om
Ya
hoo
F
inanc
e
a
t
ht
tps
:/
/f
inanc
e
.
ya
hoo.
c
om/
.
No
pr
opr
ieta
r
y
o
r
r
e
s
tr
icte
d
-
a
c
c
e
s
s
da
ta
we
r
e
us
e
d
in
t
his
wor
k.
RE
F
E
RE
NC
E
S
[
1]
G
.
S
onka
vde
,
D
.
S
.
D
ha
r
r
a
o,
A
.
M
.
B
onga
le
,
S
.
T
.
D
e
oka
te
,
D
.
D
or
e
s
w
a
my,
a
nd
S
.
K
.
B
ha
t,
“
F
or
e
c
a
s
ti
ng
s
to
c
k
ma
r
ke
t
pr
ic
e
s
us
in
g
ma
c
hi
ne
le
a
r
ni
ng
a
nd
de
e
p
le
a
r
ni
ng
mode
ls
:
a
s
ys
te
ma
t
ic
r
e
vi
e
w
,
pe
r
f
or
ma
nc
e
a
na
l
ys
is
a
nd
di
s
c
us
s
io
n
of
im
pl
ic
a
ti
o
ns
,”
I
nt
e
r
nat
io
nal
J
our
nal
of
F
in
anc
ia
l
St
udi
e
s
, vol
. 11, no. 3, 2023, doi
:
10.3390/i
jf
s
11030094.
[
2]
C
.
Y
.
L
in
a
nd
J
.
A
.
L
.
M
a
r
que
s
,
“
S
to
c
k
ma
r
ke
t
pr
e
di
c
ti
on
us
in
g
a
r
ti
f
ic
ia
l
in
te
ll
ig
e
nc
e
:
a
s
y
s
te
ma
ti
c
r
e
vi
e
w
of
s
y
s
te
ma
ti
c
r
e
vi
e
w
s
,”
Soc
ia
l
Sc
ie
nc
e
s
and Hu
m
ani
ti
e
s
O
pe
n
, vol
. 9, 2024, doi:
10.101
6/
j.
s
s
a
ho.2024.100864.
[
3]
M
.
A
.
A
.
B
a
ll
e
s
te
r
os
a
nd
E
.
A
.
M
.
M
ir
a
nda
,
“
S
to
c
k
ma
r
ke
t
f
or
e
c
a
s
ti
ng
us
in
g
a
ne
ur
a
l
ne
twor
k
th
r
ough
f
unda
me
nt
a
l
in
d
ic
a
to
r
s
,
te
c
hni
c
a
l
in
di
c
a
to
r
s
a
nd ma
r
ke
t
s
e
nt
im
e
nt
a
na
ly
s
i
s
,”
C
om
put
at
i
onal
E
c
onomic
s
, 2024, doi:
10.1007/s
10614
-
024
-
10711
-
4.
[
4]
V
.
Z
a
ka
mul
in
a
nd
J
.
G
in
e
r
,
“
T
r
e
nd
f
ol
lo
w
in
g
w
it
h
mom
e
nt
um
ve
r
s
us
movi
ng
a
ve
r
a
ge
s
:
a
ta
le
of
di
f
f
e
r
e
nc
e
s
,”
Q
uant
it
at
iv
e
F
in
anc
e
, vol
. 20, no. 6, pp. 985
–
1007, 2020, doi:
10.1080/1469
7688.2020.1716057.
[
5]
S
.
S
r
iv
a
s
ta
va
,
M
.
P
a
nt
,
a
nd
V
.
G
upt
a
,
“
A
na
ly
s
is
a
nd
pr
e
di
c
ti
on
of
I
ndi
a
n
s
to
c
k
ma
r
ke
t:
a
ma
c
hi
ne
-
le
a
r
ni
ng
a
ppr
oa
c
h,”
I
nt
e
r
nat
io
nal
J
our
nal
of
Sy
s
te
m
A
s
s
ur
anc
e
E
ngi
ne
e
r
in
g
and
M
anage
m
e
nt
,
vol
.
14,
no.
4,
pp.
1567
–
1585,
2023,
doi
:
10.1007/s
13198
-
023
-
01934
-
z.
[
6]
Y
.
X
u,
D
. W
a
ng,
a
nd
J
.
H
a
o,
“
S
to
c
k
pr
ic
e
pr
e
di
c
ti
on
ba
s
e
d
on s
ha
r
e
hol
di
ng
ne
twor
k
to
pol
ogy
a
nd
L
S
T
M
mode
l
,”
P
r
oc
e
e
di
ng
s
o
f
th
e
2nd
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on
P
ubl
ic
M
anage
m
e
nt
,
D
i
gi
ta
l
E
c
onomy
and
I
nt
e
r
ne
t
T
e
c
hnol
ogy
,
I
C
P
D
I
2023
,
E
A
I
,
2
023,
doi
:
10.4108/e
a
i.
1
-
9
-
2023.2338695.
[
7]
Z
.
F
a
th
a
li
,
Z
.
K
odi
a
,
a
nd
L
.
B
.
S
a
id
,
“
S
to
c
k
ma
r
ke
t
p
r
e
di
c
ti
on
of
N
i
f
ty
50
in
de
x
a
ppl
yi
ng
ma
c
hi
ne
le
a
r
ni
ng
te
c
hni
que
s
,”
A
ppl
ie
d
A
r
ti
fi
c
ia
l
I
nt
e
ll
ig
e
nc
e
, vol
. 36, no. 1, 2022, doi
:
10.1080/08839
514.2022.2111134.
[
8]
D
.
M
uha
mm
a
d,
I
.
A
hme
d,
K
.
N
a
v
e
e
d,
a
nd
M
.
B
e
nd
e
c
ha
c
h
e
,
“
A
n
e
xpl
a
in
a
bl
e
de
e
p
le
a
r
ni
ng
a
ppr
oa
c
h
f
or
s
to
c
k
ma
r
ke
t
tr
e
nd
pr
e
di
c
ti
on,”
H
e
li
y
on
, vol
. 10, no. 21, 2024, doi
:
10.1016/j
.he
li
yon.2024.e
40095.
[
9]
P
.
K
uma
r
,
L
.
H
ot
a
,
V
.
A
.
T
ik
ki
w
a
l,
a
nd
A
.
K
uma
r
,
“
A
na
l
y
s
in
g
f
or
e
c
a
s
ti
ng
of
s
to
c
k
pr
ic
e
s
:
a
n
e
xpl
a
in
a
bl
e
A
I
A
ppr
oa
c
h,”
P
r
oc
e
di
a C
om
put
e
r
S
c
ie
nc
e
, vol
. 235, pp. 2009
–
2016, 2024, do
i:
10.1016/j
.pr
oc
s
.2024.04.190.
[
10]
M
.
T
.
H
os
a
in
,
J
.
R
.
J
im
,
M
.
F
.
M
r
id
ha
,
a
nd
M
.
M
.
K
a
bi
r
,
“
E
xpl
a
in
a
bl
e
A
I
a
ppr
oa
c
he
s
in
de
e
p
le
a
r
n
in
g:
a
dva
nc
e
me
nt
s
,
a
ppl
ic
a
ti
ons
a
nd c
ha
ll
e
ng
e
s
,”
C
om
put
e
r
s
and E
le
c
tr
ic
al
E
ngi
ne
e
r
in
g
, vol
. 117, 2024, doi:
10.1016/j
.c
ompe
le
c
e
ng.2024.109246
.
Evaluation Warning : The document was created with Spire.PDF for Python.