I
n
t
e
r
n
at
ion
al
Jou
r
n
al
of
E
lec
t
r
ical
an
d
Com
p
u
t
e
r
E
n
gin
e
e
r
in
g
(
I
JE
CE
)
Vol.
1
4
,
No.
5
,
Oc
tober
20
2
4
,
pp
.
5502
~
5512
I
S
S
N:
2088
-
8708
,
DO
I
:
10
.
11591/i
jec
e
.
v
1
4
i
5
.
pp
5
502
-
5512
5502
Jou
r
n
al
h
omepage
:
ht
tp:
//
ij
e
c
e
.
iaes
c
or
e
.
c
om
Pr
e
d
ic
t
i
n
g t
e
l
e
vi
si
on
p
r
ogr
am
s su
c
c
e
ss u
si
n
g m
a
c
h
i
n
e
l
e
ar
n
in
g
t
e
c
h
n
i
q
u
e
s
Kh
ali
d
E
l
F
ayq
,
S
ai
d
T
k
at
e
k
,
L
ah
c
e
n
I
d
ou
gl
id
C
omput
e
r
S
c
ie
nc
e
s
R
e
s
e
a
r
c
h L
a
bor
a
to
r
y, F
a
c
ul
ty
of
S
c
ie
nc
e
, I
bn T
of
a
il
U
ni
ve
r
s
it
y, K
e
ni
tr
a
, M
or
oc
c
o
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
Aug
15,
2023
R
e
vis
e
d
J
un
17,
2024
Ac
c
e
pted
J
ul
2,
2024
In
t
h
e
e
v
er
-
ev
o
l
v
i
n
g
me
d
i
a
l
a
n
d
s
cap
e,
t
el
e
v
i
s
i
o
n
(T
V
)
remai
n
s
a
c
o
v
e
t
ed
p
l
a
t
fo
rm,
c
o
mp
e
l
l
i
n
g
i
n
d
u
s
t
ry
p
l
ay
er
s
t
o
i
n
n
o
v
a
t
e
am
i
d
i
n
t
en
s
e
co
m
p
et
i
t
i
o
n
.
T
h
i
s
s
t
u
d
y
fo
c
u
s
e
s
o
n
l
e
v
erag
i
n
g
mach
i
n
e
l
ear
n
i
n
g
r
eg
res
s
i
o
n
mo
d
el
s
t
o
p
reci
s
el
y
p
red
i
ct
TV
p
r
o
g
r
am
reach
.
O
u
r
o
b
j
ect
i
v
e
i
s
t
o
as
s
e
s
s
t
h
e
mo
d
el
s
'
effi
cacy
,
rev
eal
i
n
g
a
s
t
an
d
o
u
t
p
erfo
rmer
w
i
t
h
a
mean
a
b
s
o
l
u
t
e
p
ercen
t
erro
r
o
f
j
u
s
t
u
n
d
er
8
%
.
Si
g
n
i
fi
ca
n
t
l
y
,
w
e
i
d
e
n
t
i
fy
feat
u
res
e
x
ert
i
n
g
a
s
u
b
s
t
an
t
i
al
i
mp
ac
t
o
n
p
red
i
ct
i
o
n
s
a
n
d
e
x
p
l
o
re
t
h
e
p
o
t
en
t
i
a
l
fo
r
mo
d
e
l
en
h
a
n
ceme
n
t
t
h
r
o
u
g
h
e
x
p
a
n
d
e
d
d
a
t
as
e
t
s
.
T
h
i
s
r
es
earc
h
ex
t
en
d
s
b
ey
o
n
d
s
t
at
i
s
t
i
ca
l
i
n
s
i
g
h
t
s
,
o
fferi
n
g
act
i
o
n
ab
l
e
i
mp
l
i
ca
t
i
o
n
s
fo
r
T
V
ch
an
n
el
man
a
g
ers
.
E
mp
o
w
ere
d
b
y
t
h
e
s
e
fi
n
d
i
n
g
s
,
man
ag
er
s
can
mak
e
i
n
fo
rme
d
d
eci
s
i
o
n
s
i
n
p
r
o
g
ram
p
l
an
n
i
n
g
an
d
s
c
h
e
d
u
l
i
n
g
,
o
p
t
i
mi
z
i
n
g
v
i
ew
er
e
n
g
a
g
emen
t
.
T
h
e
t
emp
o
ra
l
an
a
l
y
s
i
s
o
f
ev
o
l
v
i
n
g
t
ren
d
s
o
v
er
t
i
me
ad
d
s
a
n
u
an
ce
d
l
ay
er
t
o
o
u
r
s
t
u
d
y
,
al
i
g
n
i
n
g
i
t
w
i
t
h
t
h
e
d
y
n
am
i
c
n
a
t
u
re
o
f
t
h
e
me
d
i
a
l
an
d
s
ca
p
e.
A
s
t
el
e
v
i
s
i
o
n
ret
ai
n
s
i
t
s
d
y
n
ami
c
fo
rce,
o
u
r
i
n
s
i
g
h
t
s
co
n
t
r
i
b
u
t
e
n
o
t
o
n
l
y
t
o
acad
em
i
c
d
i
s
co
u
rs
e
b
u
t
al
s
o
p
ro
v
i
d
e
p
ract
i
cal
g
u
i
d
an
ce,
en
h
an
ci
n
g
t
h
e
co
m
p
et
i
t
i
v
e
ed
g
e
o
f
t
e
l
ev
i
s
i
o
n
ch
an
n
el
s
.
K
e
y
w
o
r
d
s
:
Da
ta
Da
ta
s
c
ienc
e
F
or
e
c
a
s
t
models
M
a
c
hine
lea
r
ning
P
r
e
dicting
T
e
levis
ion
T
e
levis
ion
pr
ogr
a
m
Th
i
s
i
s
a
n
o
p
en
a
c
ces
s
a
r
t
i
c
l
e
u
n
d
e
r
t
h
e
CC
B
Y
-
SA
l
i
ce
n
s
e.
C
or
r
e
s
pon
din
g
A
u
th
or
:
Kha
li
d
E
l
F
a
yq
C
omput
e
r
S
c
ienc
e
s
R
e
s
e
a
r
c
h
L
a
bor
a
tor
y,
F
a
c
ult
y
of
S
c
ienc
e
,
I
bn
T
of
a
il
Unive
r
s
it
y
Ke
nit
r
a
,
M
or
oc
c
o
E
mail:
kha
li
de
lf
a
yq@gmail
.
c
om
1.
I
NT
RODU
C
T
I
ON
T
he
mar
ke
ti
ng
indus
tr
y
s
tands
a
s
one
of
the
lar
ge
s
t
in
the
wor
ld.
T
e
levis
ion
(
T
V
)
c
ompanie
s
inves
t
mi
ll
ions
of
do
ll
a
r
s
in
mar
ke
ti
ng,
a
nd
a
udienc
e
r
a
ti
ngs
c
a
n
a
s
s
is
t
a
dve
r
ti
s
e
r
s
in
obtaining
the
in
f
or
mat
ion
they
ne
e
d
to
c
r
e
a
te
a
ds
that
mee
t
thei
r
a
udienc
e
's
ne
e
d
s
.
T
his
inf
o
r
mation
a
ls
o
pr
ovides
c
ontent
c
r
e
a
tor
s
a
nd
TV
ne
twor
ks
with
the
a
bil
it
y
to
make
im
pr
ove
ments
a
nd
plays
a
c
r
uc
ial
r
ole
in
va
r
ious
a
r
e
a
s
,
includin
g
c
ontent
pr
omot
ion
a
nd
p
r
ogr
a
mm
ing
.
I
n
M
or
oc
c
o,
the
a
udienc
e
mea
s
ur
e
ment
s
ys
tem
s
tands
a
t
the
f
or
e
f
r
ont
o
f
innovation.
T
he
c
ur
r
e
nt
a
udienc
e
mea
s
ur
e
ment
methods
in
M
or
oc
c
o
ope
r
a
te
on
the
pr
inciple
of
s
a
mpl
ing.
S
e
ve
r
a
l
ke
y
or
ga
n
iza
ti
ons
play
pivot
a
l
r
oles
in
thi
s
pr
oc
e
s
s
,
including
the
High
Author
it
y
f
o
r
Audiovis
ua
l
C
omm
unica
ti
on
(
HA
C
A)
,
whic
h
is
the
r
e
gulator
y
body
ove
r
s
e
e
ing
a
udiovi
s
ua
l
c
omm
unica
ti
on,
the
I
nter
p
r
of
e
s
s
ional
M
e
dia
Audie
nc
e
C
e
nter
(
C
I
AU
M
E
D)
,
r
e
s
pons
ibl
e
f
or
e
s
tablis
hing
a
nd
im
pleme
nti
ng
the
media
mea
s
ur
e
ment
s
y
s
tem
f
or
a
udienc
e
s
in
M
or
oc
c
o
.
Additi
ona
l
ly,
M
a
r
oc
métr
i
e
is
e
ntr
us
ted
wi
th
the
c
r
uc
ial
r
e
s
pons
ibi
li
ty
o
f
c
oll
e
c
ti
ng
a
nd
pr
oc
e
s
s
ing
media
a
udien
c
e
da
ta
in
M
or
oc
c
o
,
a
nd
the
na
ti
ona
l
r
a
dio
a
nd
tele
vis
ion
c
ompany
(
S
NR
T
)
is
a
c
ti
ve
ly
e
nga
ge
d,
ove
r
s
e
e
ing
the
mea
s
ur
e
ment
o
f
a
udienc
e
s
f
or
a
ll
M
or
oc
c
a
n
c
ha
nne
ls
a
c
r
os
s
th
e
e
nti
r
e
M
or
oc
c
a
n
ter
r
it
or
y,
e
nc
ompas
s
ing
both
ur
ba
n
a
nd
r
ur
a
l
de
mogr
a
phics
.
T
o
c
ontextua
li
z
e
thi
s
s
tudy
withi
n
the
br
oa
de
r
f
ield
of
tele
vis
ion
pr
ogr
a
mm
ing
a
nd
a
udienc
e
mea
s
ur
e
ment,
it
i
s
im
pe
r
a
ti
ve
to
a
c
knowle
dge
the
s
tate
-
of
-
the
-
a
r
t.
E
xis
ti
ng
s
tudi
e
s
,
e
xe
mpl
if
ied
by
[
1]
–
[
7]
,
pr
e
domi
na
ntl
y
f
oc
us
on
pr
e
dicting
im
pr
e
s
s
ions
f
or
s
pe
c
if
ic
c
ontent,
movi
e
s
,
s
e
r
ies
,
a
nd
T
V
r
a
ti
ngs
.
How
e
ve
r
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
P
r
e
dicting
tele
v
is
ion
pr
ogr
ams
s
uc
c
e
s
s
us
ing
mac
hine
lear
ning
tec
hniques
(
K
hali
d
E
l
F
ay
q
)
5503
a
dis
c
e
r
nibl
e
ga
p
e
mer
ge
s
in
the
a
r
e
na
o
f
pr
e
dic
ti
ng
the
ove
r
a
ll
r
e
a
c
h
of
a
diver
s
e
a
udienc
e
.
T
hi
s
pa
pe
r
e
nde
a
vor
s
to
a
ddr
e
s
s
thi
s
ga
p
by
e
mpl
oying
pr
e
dictive
a
na
lyt
ics
,
s
pe
c
if
ica
ll
y
a
im
ing
to
e
nha
nc
e
the
ove
r
a
ll
T
V
a
udienc
e
pe
r
c
e
ntage
.
Dr
a
wing
ins
ight
s
f
r
om
a
s
pe
c
ialize
d
c
ompany
in
M
or
oc
c
a
n
tele
vis
ion
a
udienc
e
s
tatis
ti
c
s
,
we
s
e
e
k
to
c
ontr
ibut
e
nove
l
pe
r
s
pe
c
ti
ve
s
to
the
e
xis
ti
ng
body
of
knowle
dge
.
L
a
a
youne
c
ha
nne
l,
one
of
the
M
or
oc
c
a
n
tele
vis
ion
c
ha
nne
ls
,
ha
s
it
s
da
il
y
T
V
guides
manua
ll
y
s
c
he
duled
by
the
s
taf
f
withi
n
the
pr
e
diction
de
pa
r
tm
e
nt.
a
pr
oc
e
s
s
that
c
a
n
be
tedious
a
nd
pr
one
t
o
human
e
r
r
or
.
At
ti
mes
,
s
hows
a
r
e
s
c
he
duled
incor
r
e
c
tl
y,
c
r
e
a
ti
ng
inac
c
ur
a
c
ies
in
the
T
V
s
c
he
dule.
T
his
may
r
e
s
ult
f
r
om
mi
s
judgm
e
nts
o
f
de
mogr
a
phic
da
ta
or
T
V
r
a
ti
ngs
.
S
uc
h
e
r
r
o
r
s
c
a
n
lea
d,
f
or
e
xa
mpl
e
,
to
incor
r
e
c
t
pr
icing
f
o
r
a
dve
r
ti
s
e
ments
or
c
ontr
a
c
t
br
e
a
c
he
s
,
r
e
s
ult
ing
in
f
inanc
ial
los
s
e
s
a
nd
de
c
r
e
a
s
e
d
view
e
r
s
hip
.
T
he
f
oll
owing
outl
ines
r
e
leva
nt
wor
ks
in
the
f
ield
of
mac
hine
lea
r
ning
s
olut
ions
f
or
pr
e
diction
a
nd
c
las
s
if
ica
ti
on
withi
n
the
domain
of
a
udiovi
s
ua
l
c
ontent
[
8]
,
[
9]
,
Ac
c
ur
a
tely
f
or
e
c
a
s
ti
ng
a
udienc
e
r
a
ti
ngs
is
pa
r
a
mount
f
or
de
ter
mi
ning
a
T
V
s
how's
r
e
ve
nue
,
o
f
f
e
r
ing
be
ne
f
it
s
li
ke
e
f
f
icie
nt
inves
tm
e
nt
planning.
De
s
pit
e
the
e
vident
ne
e
d
f
o
r
pr
e
c
is
e
T
V
pr
e
diction
[
10]
,
[
11]
,
th
is
a
r
e
a
r
e
mains
r
e
latively
unde
r
e
xplor
e
d
in
pr
a
c
ti
c
e
.
R
e
c
e
nt
a
dva
nc
e
ments
in
da
ta
mi
ning
a
nd
mac
hine
lea
r
ning
ha
ve
s
pur
r
e
d
s
tudi
e
s
to
a
pply
nove
l
tec
h
niques
to
the
T
V
r
a
ti
ngs
pr
e
diction
p
r
oblem.
Nota
bly
,
s
ince
2000
,
the
r
e
ha
s
be
e
n
a
f
oc
us
o
n
indi
vid
ua
l
r
a
ti
n
g
b
e
ha
vio
r
,
inclu
ding
a
n
a
na
lys
is
of
s
oc
ial
med
ia
p
lat
f
o
r
ms
s
uc
h
a
s
F
a
c
e
boo
k
a
n
d
T
wit
ter
,
inco
r
po
r
a
t
ing
p
r
e
dic
to
r
s
s
pe
c
if
i
c
to
T
V
pr
o
gr
a
ms
or
s
e
r
ies
,
s
uc
h
a
s
pos
t
c
o
unts
,
l
ikes
,
s
ha
r
e
s
,
a
n
d
c
om
ments
f
o
r
s
pe
c
i
f
i
e
d
pa
ge
s
[
12
]
–
[
16
]
.
I
n
the
r
e
a
lm
o
f
p
r
e
dicting
T
V
r
a
ti
ngs
,
s
tudi
e
s
uti
li
z
e
a
ggr
e
ga
ted
da
ta
f
o
r
ind
ivi
dua
ls
[
17
]
,
[
18]
,
a
nd
pe
ople
mete
r
va
lues
f
or
a
ll
hous
e
holds
[
19
]
.
V
a
r
ious
tec
hniques
a
r
e
a
ppli
e
d
to
a
dd
r
e
s
s
the
T
V
r
a
ti
ng
pr
e
diction
pr
oblem
,
including
ne
ur
a
l
ne
two
r
ks
[
20]
–
[
22]
,
de
c
is
ion
tr
e
e
s
[
23
]
,
[
24]
,
a
nd
gr
a
dient
-
boos
ti
ng
mac
hines
[
25]
.
R
idge
r
e
gr
e
s
s
ion
[
26]
,
ge
ne
ti
c
a
lgor
it
hm
[
27
]
,
[
28]
,
a
nd
li
ne
a
r
r
e
gr
e
s
s
ion
[
29]
,
a
r
e
f
r
e
que
ntl
y
e
mpl
oye
d.
Ano
ther
s
tu
dy
on
mac
hi
ne
lea
r
ning
p
r
e
dicti
on
e
xa
mi
ne
d
the
pos
s
ib
il
i
ty
of
p
r
e
dic
ti
n
g
t
he
p
opula
r
it
y
of
ne
w
mo
vies
a
nd
s
e
r
ies
[
3
0]
–
[
36
]
.
I
n
thi
s
s
tu
dy,
we
tes
ted
s
e
ve
r
a
l
di
f
f
e
r
e
n
t
pr
e
dict
ion
m
ode
ls
s
uc
h
a
s
r
a
ndo
m
f
o
r
e
s
t
(
R
F
)
,
k
-
ne
a
r
e
s
t
ne
i
ghbo
r
s
(
KN
N)
,
s
u
ppor
t
ve
c
to
r
mac
hine
(
S
VM
)
,
a
n
d
li
ne
a
r
r
e
gr
e
s
s
i
on
(
L
R
)
.
T
his
pa
pe
r
int
r
oduc
e
s
a
pa
r
a
digm
-
s
hif
ti
ng
f
o
r
e
c
a
s
ti
ng
modul
e
tailor
e
d
f
or
the
L
a
a
youne
c
ha
nne
l.
At
it
s
c
or
e
l
ies
the
in
tegr
a
ti
on
of
s
upe
r
vis
e
d
mac
hine
lea
r
ning,
a
powe
r
f
ul
tool
to
a
utom
a
te
the
identi
f
i
c
a
ti
on
of
opti
mal
br
oa
dc
a
s
t
ti
mes
.
T
r
a
ined
meticulous
ly
on
his
tor
ica
l
da
ta,
the
mac
hine
lea
r
ning
model
e
mbar
ks
on
a
n
ins
ight
f
ul
a
na
lys
is
of
pa
s
t
b
r
oa
dc
a
s
ti
ng
ha
bit
s
,
identif
ies
r
e
c
ur
r
ing
pa
tt
e
r
ns
,
a
nd
a
nti
c
ipate
s
t
he
idea
l
br
oa
dc
a
s
t
ti
mes
f
or
s
pe
c
if
ic
s
hows
.
Our
s
tudy
ma
ke
s
a
s
ubs
tantial
c
ontr
ibut
ion
to
the
f
ield
by
a
ddr
e
s
s
ing
the
c
ha
ll
e
nge
s
f
a
c
e
d
by
the
L
a
a
youne
c
ha
nne
l,
in
c
ludi
ng
manua
l
s
c
he
duli
ng
e
r
r
or
s
,
f
inanc
ial
los
s
e
s
,
a
nd
de
c
r
e
a
s
e
d
view
e
r
s
hip.
T
he
int
r
oduc
ti
on
o
f
thi
s
a
ut
omate
d
pr
ogr
a
m
r
e
pr
e
s
e
nts
a
s
igni
f
ica
nt
lea
p
f
or
w
a
r
d,
not
only
r
e
duc
ing
the
labor
-
int
e
ns
ive
na
tur
e
o
f
c
r
e
a
ti
n
g
the
T
V
guide
bu
t
a
ls
o
e
leva
ti
ng
a
c
c
ur
a
c
y
a
nd
mi
nim
izing
potential
e
r
r
or
s
a
s
s
oc
iate
d
with
manua
l
p
lanning.
T
he
r
e
mainde
r
o
f
th
is
pa
pe
r
is
or
ga
nize
d
a
s
f
oll
ows
:
s
e
c
ti
on
2
de
s
c
r
ibes
the
pr
opos
e
d
a
ppr
oa
c
h
,
including
pr
oblem
f
or
mul
a
ti
on
,
da
ta
de
s
c
r
ipt
ion
,
a
nd
T
V
br
oa
dc
a
s
t
pr
e
pa
r
a
ti
on
.
S
e
c
ti
on
3
is
de
dica
t
e
d
to
the
pr
e
s
e
ntation
a
nd
a
na
lys
is
o
f
r
e
s
ult
s
,
of
f
e
r
ing
a
nua
nc
e
d
unde
r
s
tanding
o
f
pr
e
dictive
models
f
or
top
-
r
a
ted
T
V
s
hows
.
S
e
c
ti
on
4
c
onc
ludes
with
ins
ight
s
int
o
f
utu
r
e
e
xtens
ions
a
nd
a
ppli
c
a
ti
ons
of
our
methodology.
2.
M
E
T
HO
D
T
his
c
ha
pter
will
p
r
e
s
e
nt
the
methods
a
nd
the
a
r
guments
e
mpl
oye
d
in
thi
s
s
tudy
a
long
with
the
jus
ti
f
ica
ti
ons
f
or
their
us
e
in
thi
s
pa
pe
r
.
A
c
ompr
e
he
ns
ive
li
ter
a
tur
e
r
e
view
wa
s
unde
r
take
n
to
identif
y
a
nd
a
s
s
e
s
s
va
r
ious
types
of
mac
hine
lea
r
ning
mod
e
ls
a
ppli
c
a
ble
to
the
given
p
r
oblem.
T
his
inv
olved
a
n
e
xa
mi
na
ti
on
of
pr
io
r
s
tudi
e
s
,
r
e
s
e
a
r
c
h,
a
nd
wor
k
r
e
late
d
to
s
upe
r
vis
e
d
mac
hine
lea
r
ning.
W
e
then
a
s
s
e
s
s
e
d
a
nd
c
ompar
e
d
the
c
hos
e
n
models
us
ing
diver
s
e
me
thods
.
F
inally
,
we
c
oll
e
c
ted
a
nd
a
na
lyze
d
the
r
e
s
ul
ts
of
the
e
va
luation
a
nd
c
ompar
is
on.
T
he
c
hoice
to
e
mpl
oy
a
s
upe
r
vis
e
d
mac
hine
lea
r
ni
ng
model
wa
s
gr
ounde
d
in
i
ts
c
a
pa
c
it
y
to
pr
e
dict
da
ta
ba
s
e
d
on
it
s
pr
ior
knowle
dge
o
f
the
da
tas
e
t.
E
xtens
ive
r
e
s
e
a
r
c
h
a
nd
a
na
lys
is
of
va
r
ious
pr
e
dictive
models
r
e
ve
a
led
that
s
upe
r
vis
e
d
lea
r
ning
methods
we
r
e
mor
e
s
uit
a
ble
f
o
r
the
tas
k
a
t
ha
nd
c
ompar
e
d
to
other
a
ppr
oa
c
he
s
.
F
ur
ther
mor
e
,
the
na
tur
e
of
the
da
ta
us
e
d
in
s
im
il
a
r
s
tudi
e
s
a
li
gne
d
with
the
da
ta
a
va
il
a
ble
in
thi
s
pr
ojec
t,
r
e
inf
or
c
ing
the
pr
e
f
e
r
e
nc
e
f
or
s
upe
r
vis
e
d
models
.
B
a
s
e
d
on
the
r
e
s
e
a
r
c
h
r
e
view
e
d,
we
de
c
ided
to
uti
li
z
e
k
-
ne
a
r
e
s
t
ne
ighbor
s
r
e
gr
e
s
s
ion,
de
c
is
ion
tr
e
e
,
r
a
ndom
f
o
r
e
s
t,
li
ne
a
r
r
e
g
r
e
s
s
ion,
s
uppor
t
ve
c
tor
m
a
c
hine,
mul
ti
-
laye
r
pe
r
c
e
ptr
on
(
M
L
P
)
ne
ur
a
l
ne
tw
or
k
,
a
nd
r
idge
r
e
gr
e
s
s
ion
.
T
he
s
e
models
we
r
e
de
e
med
mos
t
a
ppr
opr
iate
ba
s
e
d
on
their
pe
r
f
or
manc
e
a
nd
a
ppl
ica
bil
it
y
in
pr
e
vious
s
tudi
e
s
.
B
y
e
mpl
oying
a
c
ombi
na
ti
o
n
of
thes
e
c
hos
e
n
models
,
we
a
im
to
a
c
hieve
a
c
c
ur
a
te
pr
e
dictions
f
or
the
number
o
f
view
e
r
s
a
nd
a
udienc
e
s
ha
r
e
,
f
a
c
il
it
a
ti
ng
inf
or
med
de
c
is
ion
-
making
f
or
tele
vis
ion
pr
ogr
a
ms
.
T
he
methodology
put
f
or
wa
r
d
invol
ve
s
s
ix
s
pe
c
if
i
c
pha
s
e
s
:
da
ta
pr
e
pr
oc
e
s
s
ing,
e
xtr
a
c
ti
ng
f
e
a
tur
e
s
,
c
ons
tr
uc
ti
ng
the
pr
e
dictive
model,
a
s
s
e
s
s
ing
the
model's
pe
r
f
or
manc
e
,
a
nd
im
pleme
nti
ng
the
m
ode
l.
As
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2088
-
8708
I
nt
J
E
lec
&
C
omp
E
ng
,
Vol
.
1
4
,
No.
5
,
Oc
tober
2
02
4
:
5502
-
5512
5504
s
hown
in
F
igur
e
1,
the
ove
r
a
ll
wor
k
f
low
of
the
pr
oc
e
s
s
.
I
nit
ially,
the
r
a
w
da
ta
unde
r
goe
s
pr
e
p
r
oc
e
s
s
ing.
T
his
is
f
oll
owe
d
by
f
e
a
tur
e
e
xt
r
a
c
ti
on
to
f
a
c
il
it
a
te
the
pr
oc
e
s
s
.
S
ubs
e
que
ntl
y,
the
da
tas
e
t
is
divi
de
d
int
o
tr
a
ini
ng
a
nd
tes
t
s
e
ts
.
T
he
mac
hine
lea
r
ning
models
a
r
e
then
tr
a
ined
us
ing
the
tr
a
ini
ng
da
ta.
Af
ter
tr
a
ini
ng,
e
a
c
h
model's
pe
r
f
or
manc
e
is
a
s
s
e
s
s
e
d
thr
ough
tes
ti
ng.
F
inally,
the
model
e
xhibi
ti
ng
high
pe
r
f
o
r
manc
e
is
de
ployed
us
ing
the
F
las
k
f
r
a
mew
or
k
f
o
r
us
e
r
int
e
r
a
c
ti
on.
F
igur
e
1.
P
r
opos
e
d
model
2.
1
.
Get
t
in
g
d
at
a
r
aw
Give
n
the
im
pr
a
c
ti
c
a
li
ty
of
dir
e
c
tl
y
mea
s
ur
ing
the
tr
ue
tele
vis
ion
view
ing
r
a
te,
the
uti
li
z
e
d
T
V
r
a
ti
ng
va
lues
a
r
e
thos
e
publi
s
he
d
by
C
I
AU
M
E
D.
T
he
a
c
q
uir
e
d
da
tas
e
t
c
ompr
is
e
s
hi
s
tor
ica
l
da
ta
f
r
om
the
ye
a
r
2020,
f
e
a
tur
ing
the
a
tt
r
ibut
e
s
(
c
olum
ns
)
il
lus
tr
a
ted
in
F
i
gur
e
2.
F
ur
ther
mor
e
,
the
da
tas
e
t
is
c
a
tegor
ize
d
int
o
twe
lve
dis
ti
nc
t
c
las
s
e
s
,
a
s
de
picte
d
in
F
igu
r
e
3.
T
he
a
udienc
e
mea
s
ur
e
ment
or
ga
niza
ti
on
in
M
or
oc
c
o
us
e
s
a
s
tatis
ti
c
a
l
method
c
a
ll
e
d
“
M
e
diame
tr
ie
”
to
a
s
s
e
s
s
tele
vi
s
ion
c
ha
nne
l
view
e
r
s
hip.
T
his
tec
hnique
r
e
li
e
s
on
a
c
a
r
e
f
ull
y
s
e
lec
ted
pa
ne
l
of
in
divi
dua
ls
r
e
pr
e
s
e
nti
ng
the
c
ountr
y's
tele
vis
ion
a
udienc
e
.
T
h
e
s
e
pa
ne
li
s
ts
a
r
e
e
quipped
with
mea
s
ur
e
ment
de
vice
s
,
s
uc
h
a
s
boxe
s
or
s
of
twa
r
e
,
whic
h
a
utom
a
ti
c
a
ll
y
r
e
c
or
d
the
pr
ogr
a
ms
they
wa
tch
on
tele
vis
ion.
A
c
omm
on
pr
a
c
ti
c
e
in
a
udienc
e
mea
s
ur
e
ment
is
the
us
e
of
inaudible
a
udio
c
ode
s
,
a
ls
o
known
a
s
a
udio
identif
ica
ti
on
c
od
e
s
(
AI
C
)
or
a
udio
wa
ter
mar
king
c
ode
s
.
I
ns
e
r
ted
int
o
the
b
r
oa
dc
a
s
t
s
ignal
of
tele
vis
ion
c
ha
nne
ls
whic
h
a
r
e
s
hown
in
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
P
r
e
dicting
tele
v
is
ion
pr
ogr
ams
s
uc
c
e
s
s
us
ing
mac
hine
lear
ning
tec
hniques
(
K
hali
d
E
l
F
ay
q
)
5505
F
igur
e
4
f
r
om
the
c
ont
r
ol
r
oo
m
of
S
NR
T
,
thes
e
c
ode
s
a
r
e
de
s
igned
to
be
inaudible
to
view
e
r
s
,
e
ns
ur
ing
a
n
undis
tur
be
d
view
ing
e
xpe
r
ienc
e
.
AI
C
plays
a
c
r
uc
ial
r
ole
in
média
métr
ie
by
pr
ov
idi
ng
a
method
to
mar
k
a
udio
c
ontent.
W
he
n
a
pr
ogr
a
m
is
br
oa
dc
a
s
t,
the
c
or
r
e
s
ponding
a
udio
identif
ica
ti
on
c
ode
is
a
ls
o
tr
a
ns
mi
tt
e
d
with
the
c
ontent.
M
e
a
s
ur
e
ment
de
vice
s
with
view
e
r
s
a
r
e
de
s
igned
to
de
tec
t
thes
e
inaudible
c
ode
s
.
W
he
n
a
c
ode
is
de
tec
ted,
inf
or
mation
a
bout
the
wa
tche
d
pr
og
r
a
m
is
r
e
c
or
de
d
a
nd
tr
a
ns
mi
tt
e
d
to
the
mea
s
ur
e
ment
or
ga
niza
ti
on.
T
hus
,
the
da
ta
c
ol
lec
ted
f
r
om
the
r
e
pr
e
s
e
ntative
pa
ne
l
a
ll
ows
C
I
AU
M
E
D
to
e
s
ti
mate
the
tot
a
l
nu
mber
of
view
e
r
s
wa
tching
e
a
c
h
c
ha
nne
l
a
t
a
ny
given
t
im
e
,
a
long
wi
th
their
de
mog
r
a
phic
pr
o
f
il
e
.
F
igur
e
2.
E
xa
mpl
e
of
da
ta
f
r
om
C
I
AU
M
E
D
F
igur
e
3.
Ge
nr
e
dis
tr
ibut
ion
o
f
T
V
pr
og
r
a
ms
F
igur
e
4.
Audie
nc
e
mea
s
ur
e
ment
e
quipm
e
nt
in
the
c
ontr
ol
r
oom
(
S
NR
T
)
T
he
ge
ne
r
a
ti
on
o
f
a
udio
identif
ica
ti
on
c
ode
s
us
e
s
va
r
ious
tec
hniques
,
including
pha
s
e
modul
a
ti
on.
T
his
method
invo
lves
s
ubtl
e
va
r
iations
in
the
pha
s
e
of
the
o
r
igi
na
l
a
udio
s
ignal,
making
the
c
ode
s
i
na
udibl
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2088
-
8708
I
nt
J
E
lec
&
C
omp
E
ng
,
Vol
.
1
4
,
No.
5
,
Oc
tober
2
02
4
:
5502
-
5512
5506
to
human
li
s
tene
r
s
but
de
tec
table
by
mea
s
ur
e
ment
or
tr
a
c
king
de
vice
s
.
Othe
r
a
ppr
oa
c
he
s
,
s
uc
h
a
s
F
our
ier
tr
a
ns
f
or
mation,
f
r
e
que
nc
y
modul
a
ti
on,
a
nd
ps
e
udo
-
r
a
ndom
c
ode
s
,
a
r
e
a
ls
o
e
mpl
oye
d
to
c
r
e
a
te
AI
C
.
T
he
s
e
a
udio
identif
ica
ti
on
c
ode
s
a
r
e
de
s
igned
to
be
r
obus
t
a
ga
ins
t
c
omm
on
a
udio
dis
tor
ti
ons
a
nd
pr
oc
e
s
s
ing,
e
ns
ur
ing
a
c
c
ur
a
te
de
tec
ti
on
e
ve
n
in
c
a
s
e
s
of
a
lt
e
r
e
d
tr
a
ns
mi
s
s
ion
or
r
e
p
r
oduc
ti
on
of
the
a
udio
s
ignal.
2.
2
.
Dat
a
f
e
at
u
r
e
s
an
d
f
or
m
a
t
t
in
g
B
e
f
or
e
buil
ding
a
s
c
or
e
p
r
e
diction
model
,
da
ta
p
r
e
pr
oc
e
s
s
ing
is
a
c
r
it
ica
l
a
nd
f
unda
menta
l
s
tep
in
the
da
ta
a
na
lys
is
a
nd
mac
hine
lea
r
ning
p
ipeline.
I
t
in
volves
tr
a
ns
f
or
mi
ng
r
a
w
da
ta
in
to
a
c
lea
n,
or
ga
ni
z
e
d,
a
nd
s
tr
uc
tur
e
d
f
or
mat
that
is
s
uit
a
ble
f
or
a
na
lys
is
a
nd
modeling.
Dur
ing
thi
s
s
tage
,
va
r
ious
tec
hniques
a
r
e
a
ppli
e
d
to
ha
ndle
m
is
s
ing
va
lues
,
outl
ier
s
,
a
nd
nois
e
,
e
ns
ur
ing
da
ta
qua
li
ty
a
nd
r
e
li
a
bil
it
y
.
Addit
ionally,
da
ta
nor
maliza
ti
on
a
nd
s
c
a
li
ng
may
be
pe
r
f
or
med
to
br
ing
a
ll
f
e
a
tur
e
s
to
a
c
omm
on
s
c
a
le,
pr
e
ve
nti
ng
a
ny
f
e
a
tur
e
f
r
om
domi
na
ti
ng
the
o
ther
s
.
C
a
tegor
ica
l
va
r
iable
s
mi
ght
be
e
nc
ode
d
int
o
numer
ic
r
e
pr
e
s
e
ntations
,
e
na
bli
ng
a
lgor
it
hms
to
pr
oc
e
s
s
them
e
f
f
e
c
ti
ve
ly.
Da
ta
pr
e
p
r
oc
e
s
s
ing
s
e
t
s
the
s
tage
f
or
a
c
c
ur
a
te
f
e
a
tur
e
e
xtr
a
c
ti
on
a
nd
model
buil
ding,
e
nha
nc
ing
the
ove
r
a
ll
pe
r
f
o
r
m
a
nc
e
a
nd
int
e
r
pr
e
tabili
ty
of
the
pr
e
dictive
mod
e
ls
.
T
he
a
c
quir
e
d
da
ta
s
e
t
c
ontains
many
va
r
iable
s
,
a
nd
in
o
ur
s
tudy,
we
ne
e
d:
-
P
r
ogr
a
m
R
e
f
:
thi
s
c
ode
is
us
e
d
to
uniquely
identi
f
y
a
nd
r
e
f
e
r
e
nc
e
e
a
c
h
p
r
ogr
a
m.
˗
P
r
ogr
a
m:
T
he
na
me
o
f
the
T
V
pr
ogr
a
m
˗
Ge
nr
e
:
Ge
nr
e
of
the
T
V
s
how
˗
Numbe
r
of
view
e
r
s
:
T
he
numbe
r
o
f
view
e
r
s
f
o
r
e
a
c
h
T
V
p
r
ogr
a
m
˗
Audie
nc
e
r
a
te
(
%
)
:
Vie
we
r
s
hip
r
a
ti
ng
c
ompar
e
d
to
other
T
V
pr
og
r
a
ms
on
the
other
M
or
oc
c
a
n
T
V
s
tat
ion
˗
S
tar
t
ti
me:
T
he
s
tar
t
ti
me
of
the
br
oa
dc
a
s
t
˗
B
r
oa
dc
a
s
t
da
te
:
T
he
da
te
of
pr
ogr
a
m
br
oa
dc
a
s
ti
ng
˗
Dur
a
ti
on:
T
he
dur
a
ti
on
of
the
T
V
pr
ogr
a
m
in
m
inu
tes
˗
P
ubli
c
_Holiday:
T
his
c
olum
n
is
us
e
d
to
mar
k
i
f
a
T
V
pr
og
r
a
m
wa
s
br
oa
dc
a
s
t
on
a
pub
li
c
holi
da
y.
T
he
da
tas
e
ts
il
lus
tr
a
te
the
s
tar
t
ti
me
a
nd
da
te
of
a
pr
ogr
a
m
withi
n
a
da
y
,
a
s
s
hown
in
F
igur
e
5(
a
)
a
nd
5
(
b)
.
T
he
s
e
f
igur
e
s
s
how
two
pa
r
ts
of
the
s
a
me
table
,
s
p
li
t
f
or
c
lar
it
y
.
C
ha
nge
s
to
the
da
ta
invol
ve
d
e
li
mi
n
a
ti
ng
“
:”
to
a
void
c
onf
us
ion
with
"
/"
,
f
a
c
il
it
a
ti
ng
the
tr
a
ns
f
or
mation
o
f
text
int
o
numer
ica
l
va
lues
.
T
his
met
hod
wa
s
s
im
il
a
r
ly
e
mpl
oye
d
in
c
a
tegor
izing
the
da
tas
e
t
ba
s
e
d
on
T
V
pr
ogr
a
m
c
a
tegor
ies
.
(
a
)
(
b)
F
igur
e
5.
Da
ta
us
e
d
f
o
r
pr
e
diction
models
(
a
)
f
i
r
s
t
ha
lf
a
nd
(
b
)
s
e
c
ond
ha
lf
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
P
r
e
dicting
tele
v
is
ion
pr
ogr
ams
s
uc
c
e
s
s
us
ing
mac
hine
lear
ning
tec
hniques
(
K
hali
d
E
l
F
ay
q
)
5507
2.
3
.
F
e
a
t
u
r
e
e
n
gin
e
e
r
in
g
A
f
t
e
r
th
e
da
ta
u
nd
e
r
g
o
e
s
c
l
e
a
ni
ng
a
nd
to
k
e
ni
z
a
ti
on,
it
b
e
c
om
e
s
c
r
uc
i
a
l
to
e
xtr
a
c
t
f
e
a
t
ur
e
s
f
r
o
m
t
hi
s
pr
o
c
e
s
s
e
d
d
a
ta.
T
h
i
s
is
b
e
c
a
u
s
e
ma
c
h
in
e
s
a
r
e
in
c
a
p
a
b
l
e
of
c
o
mpr
e
he
nd
in
g
wo
r
d
s
in
t
he
ir
r
a
w
f
o
r
m
a
nd
c
a
n
on
ly
i
nt
e
r
pr
e
t
nu
m
e
r
i
c
a
l
v
a
l
ue
s
.
F
e
a
tu
r
e
e
n
gi
n
e
e
r
in
g
s
ta
nd
s
a
s
a
vi
t
a
l
s
t
e
p
i
n
th
e
d
a
t
a
pr
e
pr
o
c
e
s
s
i
ng
s
e
q
ue
n
c
e
,
f
oc
u
s
i
ng
o
n
t
he
ge
ne
r
a
ti
on
o
f
n
e
w
f
e
a
t
ur
e
s
or
th
e
m
od
if
i
c
a
ti
on
of
c
ur
r
e
n
t
on
e
s
.
T
h
i
s
s
t
e
p
i
s
e
s
s
e
nt
ia
l
f
or
im
p
r
o
vin
g
th
e
e
f
f
ic
a
c
y
of
m
a
c
hi
ne
l
e
a
r
n
in
g
a
l
gor
it
hm
s
.
T
hi
s
pr
oc
e
s
s
in
vo
lv
e
s
s
e
l
e
c
ti
ng,
t
r
a
n
s
f
or
mi
ng,
a
nd
c
r
e
a
ti
ng
f
e
a
t
ur
e
s
f
r
o
m
t
he
r
a
w
d
a
t
a
t
ha
t
b
e
s
t
c
a
p
tur
e
t
h
e
u
nd
e
r
l
yi
ng
p
a
tt
e
r
n
s
a
nd
r
e
l
a
t
io
n
s
h
ip
s
.
B
y
d
e
r
i
vi
ng
me
a
n
in
gf
ul
f
e
a
tu
r
e
s
,
t
h
e
pr
e
d
i
c
ti
v
e
p
ow
e
r
of
t
he
mo
d
e
l
s
c
a
n
b
e
s
ig
ni
f
i
c
a
nt
ly
i
mpr
ov
e
d,
l
e
a
d
in
g
t
o
be
tt
e
r
in
s
i
gh
t
s
a
n
d
a
c
c
ur
a
te
pr
e
di
c
t
io
n
s
.
I
t
r
e
q
uir
e
s
do
ma
in
kn
o
wl
e
d
ge
a
nd
a
d
e
e
p
u
n
de
r
s
ta
nd
in
g
of
t
he
da
ta
t
o
in
t
e
ll
ig
e
n
tl
y
e
ngi
n
e
e
r
f
e
a
t
ur
e
s
th
a
t
r
e
f
l
e
c
t
t
h
e
nu
a
n
c
e
s
a
nd
i
ntr
ic
a
c
ie
s
of
t
he
p
r
o
bl
e
m
do
ma
in.
2.
4
.
T
r
ain
in
g
p
r
oc
e
s
s
Af
ter
c
he
c
king
the
dis
tr
ibut
ion
of
da
ta
in
F
igur
e
6
,
e
a
c
h
e
leme
nt
with
it
s
f
r
e
que
nc
y,
a
tr
a
ini
ng
s
pli
t
of
80/20
wa
s
us
e
d
in
the
tr
a
ini
ng
pr
oc
e
s
s
.
80
%
of
t
he
da
ta
in
the
da
tas
e
t
wa
s
us
e
d
f
or
t
r
a
ini
ng
a
nd
20
%
o
f
the
da
ta
in
the
da
tas
e
t
wa
s
us
e
d
f
or
tes
ti
ng
a
f
ter
tr
a
i
ning
the
model.
T
he
de
c
is
ion
to
a
dopt
the
80/20
s
pli
t
wa
s
ba
s
e
d
on
the
s
upe
r
ior
r
e
s
ult
s
it
de
mons
tr
a
ted
c
om
pa
r
e
d
to
other
tes
ted
s
pli
ts
.
T
o
e
ns
ur
e
the
r
obus
tnes
s
of
our
a
ppr
oa
c
h,
we
c
onduc
ted
mul
ti
ple
tr
a
ini
ng
it
e
r
a
ti
o
ns
f
or
e
a
c
h
model,
s
ys
tema
ti
c
a
ll
y
e
xa
mi
ning
if
th
e
r
e
we
r
e
a
ny
va
r
iations
in
p
r
e
diction
outcome
s
.
F
igur
e
6.
P
lot
vis
ua
li
z
a
ti
on
f
o
r
c
he
c
king
the
dis
tr
ib
uti
on
of
da
ta
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2088
-
8708
I
nt
J
E
lec
&
C
omp
E
ng
,
Vol
.
1
4
,
No.
5
,
Oc
tober
2
02
4
:
5502
-
5512
5508
2.
5
.
T
e
s
t
in
g
p
r
oc
e
s
s
T
he
tes
ti
ng
pr
oc
e
s
s
pr
im
a
r
il
y
f
oc
us
e
d
on
pr
e
dicti
on
a
c
c
ur
a
c
y
to
a
s
s
e
s
s
the
r
e
lative
pe
r
f
or
manc
e
of
the
models
.
T
he
pr
im
a
r
y
objec
ti
ve
wa
s
to
e
xa
mi
ne
how
we
ll
the
models
c
ompar
e
d
in
ter
ms
of
p
r
e
diction
a
c
c
ur
a
c
y.
T
he
model
de
mons
tr
a
ti
ng
the
be
s
t
p
r
e
di
c
ti
ve
a
c
c
ur
a
c
y
wa
s
c
a
r
e
f
ull
y
s
e
lec
ted
f
or
f
ur
ther
a
na
lys
is
.
T
he
model
s
e
lec
ti
on
c
r
it
e
r
ion
wa
s
ba
s
e
d
on
pr
e
dictive
a
c
c
ur
a
c
y,
a
nd
the
be
s
t
-
pe
r
f
or
mi
ng
model
a
mon
g
other
s
wa
s
e
va
luate
d
us
ing
the
mea
n
s
qua
r
e
d
e
r
r
or
(
M
S
E
)
.
T
he
us
e
o
f
M
S
E
a
ll
ows
f
or
a
qua
nti
tative
a
n
d
r
obus
t
e
va
luation
of
model
pe
r
f
or
manc
e
,
taking
int
o
a
c
c
ount
the
a
c
c
ur
a
c
y
of
pr
e
dictions
a
c
r
os
s
the
e
nti
r
e
da
tas
e
t.
T
he
ult
im
a
te
goa
l
of
thi
s
model
s
e
lec
ti
on
pr
oc
e
s
s
is
to
f
ind
a
model
that
e
xc
e
ls
not
only
on
tr
a
ini
ng
da
ta
but
a
ls
o
ge
ne
r
a
li
z
e
s
e
f
f
e
c
ti
ve
ly
to
make
a
c
c
ur
a
te
pr
e
dictions
on
r
e
a
l
-
wor
ld
da
ta.
T
he
ke
y
s
teps
in
the
model
s
e
lec
ti
on
pr
oc
e
s
s
in
thi
s
pr
ojec
t
c
a
n
be
s
umm
a
r
ize
d
a
s
f
oll
ows
:
2.
5.
1
.
M
od
e
l
e
val
u
at
ion
T
o
c
omm
e
nc
e
the
model
e
va
luation
pr
oc
e
s
s
,
we
s
tar
t
by
c
ons
ider
ing
a
va
r
iety
o
f
c
a
ndidate
a
lgor
it
hms
.
T
he
s
e
include
li
ne
a
r
r
e
g
r
e
s
s
ion,
r
a
n
dom
f
o
r
e
s
ts
,
de
c
is
ion
tr
e
e
s
,
R
idge,
E
las
ti
c
Ne
t,
k
-
ne
a
r
e
s
t
ne
ighbor
s
,
a
nd
ne
ur
a
l
ne
two
r
ks
.
I
t
is
c
r
uc
ial
to
no
te
that
e
a
c
h
a
lgor
it
hm
c
omes
with
it
s
own
s
e
t
of
s
tr
e
ngths
a
nd
we
a
kne
s
s
e
s
,
making
it
i
mper
a
ti
ve
to
c
ompr
e
he
ns
ively
tes
t
mul
ti
ple
opti
ons
.
2.
5.
2
.
Hyp
e
r
p
ar
am
e
t
e
r
t
u
n
i
n
g
an
d
c
r
os
s
-
vali
d
at
ion
I
n
the
c
ontext
of
the
tele
vis
ion
pr
og
r
a
m
p
r
e
diction
pr
ojec
t,
a
n
a
dva
nc
e
d
a
nd
r
obus
t
e
va
luation
s
tr
a
tegy
is
im
pleme
nted
by
c
ombi
ning
hype
r
pa
r
a
mete
r
tuni
ng
with
c
r
os
s
-
va
li
da
ti
on.
T
his
dua
l
a
ppr
oa
c
h
e
ns
ur
e
s
a
thor
ough
e
xplor
a
ti
on
o
f
model
pe
r
f
or
m
a
nc
e
.
T
he
c
r
uc
ial
s
teps
invol
ve
d
in
hype
r
pa
r
a
met
e
r
tuni
ng
e
na
ble
the
identif
ica
ti
on
of
opti
mal
model
c
onf
ig
ur
a
ti
ons
,
while
c
r
os
s
-
va
li
da
ti
on
e
nha
nc
e
s
the
r
e
li
a
bil
it
y
of
the
r
e
s
ult
s
by
tes
ti
ng
the
models
a
c
r
os
s
va
r
ious
s
ubs
e
ts
of
the
da
tas
e
t.
T
hr
ough
thes
e
c
a
r
e
f
ull
y
o
r
c
he
s
tr
a
ted
s
teps
,
we
a
im
to
pinpoi
nt
the
mos
t
a
c
c
ur
a
te
a
nd
r
e
l
iable
pr
e
dictive
models
,
a
c
r
it
ica
l
f
a
c
tor
c
ontr
ibut
i
ng
to
the
ove
r
a
ll
s
uc
c
e
s
s
of
the
pr
ojec
t.
2.
5.
3
.
Com
p
ar
is
on
an
d
s
e
lec
t
ion
B
y
c
ombi
ning
hype
r
pa
r
a
mete
r
tuni
ng
with
c
r
os
s
-
va
li
da
ti
on,
a
mor
e
c
ompr
e
he
ns
ive
a
nd
r
igor
ous
a
ppr
oa
c
h
is
a
c
hieve
d
to
e
va
luate
the
pe
r
f
or
manc
e
of
mac
hine
lea
r
n
ing
models
in
the
tele
vis
ion
pr
ogr
a
m
pr
e
diction
p
r
ojec
t.
T
he
s
e
s
teps
he
lp
identif
y
the
be
s
t
model
c
onf
igur
a
ti
ons
a
nd
hype
r
pa
r
a
mete
r
s
to
o
btain
the
mos
t
a
c
c
ur
a
te
a
nd
r
e
li
a
ble
pr
e
dictions
,
c
ontr
ibu
ti
ng
to
the
ove
r
a
ll
s
uc
c
e
s
s
of
the
pr
ojec
t.
Af
t
e
r
us
ing
a
ppr
opr
iate
pe
r
f
or
manc
e
metr
ics
f
or
pr
e
diction
tas
ks
,
s
uc
h
a
s
mea
n
s
qua
r
e
d
e
r
r
or
(
M
S
E
)
,
r
oot
mea
n
s
qua
r
e
d
e
r
r
or
(
R
M
S
E
)
,
a
nd
mea
n
a
bs
olut
e
e
r
r
or
(
M
AE
)
,
w
e
c
ompar
e
d
the
pe
r
f
or
manc
e
of
di
f
f
e
r
e
nt
models
b
a
s
e
d
on
the
c
hos
e
n
metr
ics
.
I
n
ou
r
a
na
lys
is
,
we
c
a
r
e
f
u
ll
y
c
ons
ider
e
d
the
t
r
a
de
-
of
f
be
twe
e
n
bias
a
nd
va
r
ianc
e
,
a
im
ing
to
s
e
lec
t
a
model
that
mi
nim
ize
s
ove
r
f
it
ti
ng
while
c
a
ptur
ing
unde
r
lyi
ng
pa
tt
e
r
ns
.
Onc
e
the
opti
mal
m
ode
l
wa
s
identif
ied,
it
wa
s
tr
a
ined
on
the
e
nti
r
e
da
tas
e
t,
a
nd
pr
e
dictions
we
r
e
made
on
ne
w
da
ta
to
va
l
idate
it
s
e
f
f
e
c
ti
ve
ne
s
s
.
2.
6
.
T
ools
an
d
f
r
am
e
wor
k
I
n
the
c
ontext
o
f
a
p
r
e
diction
p
r
ojec
t,
a
diver
s
e
s
e
t
of
tool
s
a
nd
f
r
a
mew
or
ks
is
c
omm
on
ly
e
mpl
oye
d
to
e
xe
c
ute
e
s
s
e
nti
a
l
tas
k
s
,
including
da
ta
manipulation,
the
de
ve
lopm
e
nt
o
f
pr
e
dictive
models
,
a
nd
the
e
va
luation
of
r
e
s
ult
s
.
S
ome
popular
tool
s
a
nd
f
r
a
mew
or
ks
include
s
c
iki
t
-
lea
r
n,
P
a
nda
s
,
NumP
y,
a
nd
da
ta
vis
ua
li
z
a
ti
on
li
br
a
r
ies
li
ke
M
a
tpl
otl
ib
a
nd
S
e
a
bor
n
.
S
c
iki
t
-
lea
r
n
is
a
n
ope
n
-
s
our
c
e
li
br
a
r
y
that
of
f
e
r
s
tool
s
f
or
da
ta
a
na
lys
is
,
pr
e
diction,
a
nd
mac
hine
-
lea
r
ning
models
.
P
a
nda
s
,
a
nother
ope
n
-
s
our
c
e
s
of
twa
r
e
li
br
a
r
y,
pr
ovides
a
s
e
t
of
tool
s
f
or
da
ta
a
na
lys
is
a
nd
manipulation.
T
h
is
li
br
a
r
y
f
a
c
il
it
a
tes
da
ta
manipulation,
making
it
s
uit
a
ble
f
or
pr
e
pa
r
ing
da
ta
f
or
lea
r
ning
in
S
c
iki
t
-
l
e
a
r
n's
pr
e
dictive
models
.
NumP
y
,
on
the
other
ha
nd,
is
a
f
r
a
mew
or
k
f
or
s
c
ientif
ic
c
omput
ing
in
P
ython
.
I
t
e
xtends
s
uppor
t
f
or
lar
ge
,
mul
ti
-
dim
e
ns
ional
a
r
r
a
ys
a
nd
matr
ice
s
,
a
long
with
a
c
omp
r
e
he
ns
ive
c
oll
e
c
ti
on
of
high
-
leve
l
mathe
matica
l
f
unc
ti
ons
f
or
a
r
r
a
y
op
e
r
a
ti
ons
.
NumP
y
plays
a
c
r
uc
ial
r
ole
in
c
onve
r
ti
ng
“
.
c
s
v
”
da
ta
int
o
a
r
r
a
y
objec
ts
,
e
s
s
e
nti
a
l
f
or
ge
ne
r
a
ti
ng
tr
a
ini
ng
a
nd
tes
t
da
tas
e
ts
f
or
the
mac
hine
lea
r
ning
models
.
3.
RE
S
UL
T
S
AN
D
DI
S
CU
S
S
I
ON
T
his
c
ha
pter
de
li
ne
a
tes
the
r
e
s
ult
s
e
va
luation
pr
oc
e
s
s
.
W
e
c
onduc
ted
a
c
ompa
r
a
ti
ve
a
na
lys
is
of
the
models
to
identif
y
the
mos
t
e
f
f
e
c
ti
ve
one
.
S
ubs
e
que
ntl
y,
we
juxt
a
pos
e
d
the
top
-
pe
r
f
or
mi
ng
mo
de
ls
with
e
xpe
r
t
e
va
luations
to
a
s
s
e
s
s
the
a
c
c
ur
a
c
y
of
T
V
pr
e
dictions
.
I
n
the
c
our
s
e
o
f
the
e
xpe
r
im
e
nts
,
we
ini
ti
a
ll
y
tes
ted
a
ll
the
tr
a
ined
models
on
tes
t
da
ta.
F
oll
owin
g
the
identi
f
ica
ti
on
o
f
the
be
s
t
model
,
we
de
ployed
it
us
ing
the
e
s
s
a
y
gr
a
ding
we
b
a
ppli
c
a
ti
on
,
c
ons
tr
uc
ted
u
s
ing
P
ython
F
las
k.
As
s
hown
in
F
igur
e
7
our
m
ode
l
wa
s
e
nc
a
ps
ulate
d
in
a
pickle
f
il
e
,
f
a
c
il
it
a
ti
ng
s
e
a
ml
e
s
s
uti
li
z
a
ti
on
by
the
pr
e
diction
tea
m
a
t
the
tele
vis
ion
c
ha
nne
l.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
P
r
e
dicting
tele
v
is
ion
pr
ogr
ams
s
uc
c
e
s
s
us
ing
mac
hine
lear
ning
tec
hniques
(
K
hali
d
E
l
F
ay
q
)
5509
F
igur
e
7.
E
x
tr
a
c
t
r
e
s
ult
in
a
s
ys
tem
r
e
po
r
t
T
a
ble
1
pr
ovides
a
c
ompr
e
he
ns
ive
ove
r
view
of
the
pe
r
f
or
manc
e
s
c
or
e
s
f
or
va
r
ious
a
lgor
it
hms
tes
ted
in
the
s
tudy.
Nota
bly,
r
a
ndom
f
or
e
s
t
a
nd
k
-
ne
a
r
e
s
t
ne
ighbor
s
e
mer
ge
a
s
top
pe
r
f
or
mer
s
,
e
xhibi
ti
ng
the
highes
t
pe
r
f
or
manc
e
s
c
or
e
s
a
mong
the
tes
ted
a
lgor
it
hms
.
T
he
i
r
s
upe
r
ior
pe
r
f
or
manc
e
,
a
s
indi
c
a
ted
in
th
e
table
,
unde
r
s
c
or
e
s
their
e
f
f
e
c
ti
ve
ne
s
s
whe
n
c
ompar
e
d
to
other
a
lgor
it
hms
.
T
a
ble
1.
S
c
or
e
f
or
the
be
s
t
a
lgor
it
h
m
pe
r
f
or
manc
e
M
ode
l
R
M
S
E
M
A
E
R
2 S
c
or
e
R
id
ge
0.098081
0.078808
0.601777
L
in
e
a
r
r
e
gr
e
s
s
io
n
0.061159
0.041841
0.731777
KNN
0.056330
0.039984
0.935213
D
e
c
is
io
n t
r
e
e
0.060985
0.037925
0.752115
R
a
ndom f
or
e
s
t
0.044828
0.031368
0.966380
M
L
P
R
e
gr
e
s
s
or
0.066397
0.052581
0.803228
E
la
s
ti
c
N
e
t
0.089067
0.078565
0.631261
I
n
e
va
luating
mac
hine
lea
r
ning
a
lgor
it
hms
f
or
pr
e
dicting
tele
vis
ion
pr
ogr
a
ms
’
s
uc
c
e
s
s
,
our
tr
a
ini
ng
a
nd
tes
ti
ng
pr
oc
e
s
s
include
d
u
s
ing
ps
e
udor
a
ndom
s
ubs
e
ts
f
r
om
the
da
tas
e
t
to
pr
e
ve
nt
ove
r
f
it
ti
ng
a
nd
e
ns
ur
e
the
models
'
ge
ne
r
a
li
z
a
ti
on
c
a
pa
bil
it
ies
.
T
he
r
a
ndom
f
or
e
s
t
r
e
gr
e
s
s
or
a
lgor
it
hm
de
mons
tr
a
ted
s
upe
r
ior
pe
r
f
or
manc
e
on
the
tes
t
s
e
t
with
a
R
M
S
E
of
0
.
09653,
M
AE
of
0.
056342
,
a
nd
a
n
R
2
-
s
c
or
e
of
0.
8980,
indi
c
a
ti
ng
a
high
leve
l
of
pr
e
c
is
ion
in
it
s
p
r
e
dictions
.
Nota
bly,
whe
n
the
r
a
ndom
f
o
r
e
s
t
r
e
gr
e
s
s
or
wa
s
a
s
s
e
s
s
e
d
a
c
r
os
s
the
e
nti
r
e
da
tas
e
t,
it
c
onti
nue
d
to
s
how
e
x
c
e
pti
ona
l
pe
r
f
or
manc
e
,
a
c
hieving
a
n
R
M
S
E
of
0
.
044828,
whic
h
s
ugge
s
ts
c
ons
i
s
tent
a
nd
r
e
li
a
ble
pr
e
diction
s
.
T
he
R
2
s
c
or
e
o
f
0.
9663
a
ls
o
indi
c
a
tes
that
th
e
model
e
xplains
a
s
igni
f
ica
nt
pr
opor
ti
on
of
the
va
r
ianc
e
in
the
tar
ge
t
va
r
iable
,
e
mphas
izing
it
s
e
f
f
e
c
ti
ve
ne
s
s
a
nd
r
obus
t
pr
e
dictive
c
a
pa
bil
it
y.
T
o
vis
ua
ll
y
e
lu
c
i
da
te
o
u
r
a
na
l
yt
ica
l
o
ut
c
o
mes
,
F
i
gur
e
8
is
s
e
gm
e
n
ted
in
t
o
tw
o
s
u
b
f
i
gu
r
e
s
;
F
i
gu
r
e
8(
a
)
s
how
c
a
s
e
s
a
c
om
pa
r
a
ti
ve
a
s
s
e
s
s
m
e
n
t
o
f
the
a
lg
o
r
i
t
hm
s
ba
s
e
d
on
R
M
S
E
a
n
d
M
A
E
,
wh
il
e
F
i
gu
r
e
8(
b
)
de
l
ine
a
t
e
s
t
he
c
om
pa
r
is
o
n
ba
s
e
d
on
R
2
s
c
o
r
e
s
,
c
l
e
a
r
l
y
h
ig
hli
g
ht
in
g
th
e
R
F
m
ode
l
's
s
upe
r
io
r
p
e
r
f
o
r
m
a
nc
e
a
c
r
o
s
s
bo
th
m
e
tr
ics
.
T
h
is
o
bs
e
r
v
a
t
io
n
i
ni
ti
a
l
ly
s
u
gg
e
s
ts
the
e
f
f
ica
c
y
o
f
R
F
in
p
r
e
di
c
t
in
g
te
le
vi
s
i
on
p
r
og
r
a
m
s
uc
c
e
s
s
.
H
ow
e
ve
r
,
f
or
a
c
o
mp
r
e
he
ns
ive
e
va
l
ua
t
i
on
,
w
e
a
ls
o
c
ons
id
e
r
t
he
K
NN
mo
de
l
f
o
r
a
d
e
t
a
i
le
d
c
o
mp
a
r
is
on
a
ga
ins
t
e
xis
t
in
g
m
e
t
ho
ds
,
f
oc
us
i
ng
on
op
ti
m
iz
in
g
T
V
s
h
ow
s
c
he
du
l
in
g
s
t
r
a
te
gi
e
s
b
a
s
e
d
on
da
te
a
nd
ti
me
to
m
a
x
i
mi
z
e
v
ie
we
r
s
hi
p
.
T
h
is
a
p
p
r
oa
c
h
u
nde
r
s
c
o
r
e
s
o
u
r
c
om
m
it
m
e
n
t
to
le
ve
r
a
g
in
g
d
a
t
a
-
d
r
i
ve
n
ins
ig
h
ts
f
o
r
e
n
ha
nc
in
g
the
s
t
r
a
t
e
g
ic
pl
a
n
ni
ng
o
f
te
le
vi
s
i
on
pr
og
r
a
m
mi
ng
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2088
-
8708
I
nt
J
E
lec
&
C
omp
E
ng
,
Vol
.
1
4
,
No.
5
,
Oc
tober
2
02
4
:
5502
-
5512
5510
(
a
)
(
b)
F
igur
e
8.
C
ompar
a
ti
ve
a
na
lys
is
of
p
r
e
dictive
mode
ls
in
(
a
)
R
M
S
E
a
nd
M
AE
a
c
r
os
s
dif
f
e
r
e
nt
a
lgor
it
h
ms
a
nd
(
b)
R
2
s
c
or
e
c
ompar
is
on
a
mong
a
lgor
it
h
ms
4.
CONC
L
USI
ON
T
his
s
tudy
de
lved
int
o
the
potential
o
f
leve
r
a
ging
s
upe
r
vis
e
d
mac
hine
lea
r
ning
to
p
r
e
dict
opti
mal
T
V
s
hows
f
or
br
oa
dc
a
s
ti
ng,
e
nha
nc
e
view
e
r
s
hip,
a
nd
opti
mi
z
e
the
T
V
Guide
s
c
he
dule.
E
va
luation
of
va
r
ious
s
upe
r
vis
e
d
mac
hine
lea
r
ning
models
r
e
ve
a
led
that
r
a
ndom
f
or
e
s
t
a
nd
k
-
ne
a
r
e
s
t
ne
ighbor
s
r
e
gr
e
s
s
ion
e
xhibi
ted
the
mos
t
e
f
f
e
c
ti
ve
pe
r
f
o
r
manc
e
,
yielding
a
n
im
pr
e
s
s
ive
97%
a
c
c
ur
a
c
y
in
pr
e
dicting
the
s
uc
c
e
s
s
of
T
V
pr
ogr
a
ms
a
t
a
given
mom
e
nt.
Af
ter
pe
r
f
o
r
mi
n
g
r
e
gr
e
s
s
ion
a
na
lys
is
,
our
model
de
mons
tr
a
ted
s
upe
r
ior
pr
e
dictive
c
a
pa
bil
it
ies
,
pa
r
ti
c
ular
ly
whe
n
tr
a
ined
a
t
80%
a
nd
tes
ted
a
t
20
%
,
r
e
s
ult
ing
in
the
lea
s
t
M
AE
,
R
M
S
E
,
a
nd
M
S
E
va
lues
f
or
a
major
it
y
o
f
T
V
pr
og
r
a
ms
.
T
he
s
tudy
uti
l
ize
d
a
da
tas
e
t
s
pa
nning
th
r
e
e
ye
a
r
s
,
c
ompr
is
ing
jus
t
unde
r
12
,
500
r
ows
of
da
ta.
T
he
mac
hine
lea
r
ning
model's
pe
r
f
or
manc
e
mi
ght
ha
v
e
im
pr
ove
d
with
a
lar
ge
r
da
tas
e
t
f
or
tr
a
in
ing
a
nd
tes
ti
ng.
Although
the
model
in
thi
s
pa
pe
r
doe
s
not
r
e
plac
e
the
c
ur
r
e
nt
pr
oc
e
s
s
,
it
c
ould
s
e
r
ve
a
s
a
va
luable
tool
f
or
obtaining
quick
pr
e
dictions
.
W
hil
e
the
r
e
s
ult
s
a
r
e
pr
omi
s
ing,
a
lbeit
not
s
ur
pa
s
s
ing
the
a
c
c
ur
a
c
y
of
the
c
ur
r
e
nt
T
V
r
a
ti
ngs
e
s
ti
mation
pr
oc
e
s
s
,
ther
e
is
potential
f
or
mac
hine
lea
r
ning
models
to
r
e
plac
e
the
e
xis
ti
ng
pr
oc
e
dur
e
s
in
the
f
utur
e
.
F
utur
e
r
e
s
e
a
r
c
h
in
thi
s
a
r
e
a
s
hould
invol
ve
e
mpl
oying
s
igni
f
ica
ntl
y
lar
ge
r
da
tas
e
ts
to
a
s
s
e
s
s
their
im
pa
c
t
on
model
pe
r
f
o
r
manc
e
.
Additi
ona
ll
y,
e
x
plor
ing
a
b
r
oa
de
r
r
a
nge
o
f
mac
hine
lea
r
n
ing
mo
de
ls
a
nd
int
e
gr
a
ti
ng
the
s
tudy
da
tas
e
t
with
int
e
r
ne
t
T
V
s
tr
e
a
mi
ng
s
our
c
e
s
s
u
c
h
a
s
YouT
ube
a
nd
the
c
h
a
nne
l's
we
bs
it
e
c
ould
pr
ovide
ins
ight
s
int
o
f
u
r
ther
im
p
r
oving
pe
r
f
o
r
manc
e
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
P
r
e
dicting
tele
v
is
ion
pr
ogr
ams
s
uc
c
e
s
s
us
ing
mac
hine
lear
ning
tec
hniques
(
K
hali
d
E
l
F
ay
q
)
5511
I
n
f
utur
e
e
nde
a
vor
s
,
we
plan
to
int
r
oduc
e
e
nha
n
c
e
ments
thr
ough
the
a
dopti
on
o
f
mor
e
a
dva
nc
e
d
tec
hniques
,
s
pe
c
if
ica
ll
y
incor
por
a
ti
ng
hyb
r
idi
z
a
ti
on
with
ge
ne
ti
c
a
lgo
r
it
hms
.
T
his
s
tr
a
tegic
a
ppr
oa
c
h
holds
the
pr
omi
s
e
of
a
c
hieving
f
iner
opti
mi
z
a
ti
on
o
f
our
models
by
dyna
mi
c
a
ll
y
a
djus
ti
ng
pa
r
a
mete
r
s
.
T
his
,
in
tur
n,
will
pa
ve
the
wa
y
f
or
e
ve
n
mor
e
a
c
c
ur
a
te
pr
e
dict
ions
that
a
r
e
f
inely
tuned
to
the
nua
nc
e
d
dyna
mi
c
s
of
the
e
ve
r
-
e
volvi
ng
tele
vis
ion
lands
c
a
pe
.
AC
KNOWL
E
DGE
M
E
NT
T
he
a
uthor
s
would
li
ke
to
e
xp
r
e
s
s
their
s
ince
r
e
gr
a
ti
tude
f
or
the
s
upe
r
vis
ion
a
nd
r
e
s
our
c
e
s
pr
ovided
by
e
s
tee
med
ins
ti
tut
ions
,
including
the
Ge
ne
r
a
l
Dir
e
c
tor
a
te
of
the
Na
ti
ona
l
B
r
oa
dc
a
s
ti
ng
a
nd
T
e
levis
ion
C
ompany
of
M
or
oc
c
o
-
S
NR
T
(
htt
p
:
//
w
w
w
.
s
nr
t.
ma
)
a
nd
the
F
a
c
ult
y
of
S
c
ienc
e
s
a
t
I
bn
T
of
a
il
Unive
r
s
it
y
-
F
S
UI
T
(
htt
ps
:
//
fs
.
uit
.
ac
.
ma
).
RE
F
E
RE
NC
E
S
[
1]
K
.
L
e
e
,
J
.
P
a
r
k,
I
.
K
im
,
a
nd
Y
.
C
hoi
,
“
P
r
e
di
c
ti
ng
movi
e
s
uc
c
e
s
s
w
it
h
ma
c
hi
ne
l
e
a
r
ni
ng
te
c
hni
que
s
:
w
a
ys
to
im
pr
ove
a
c
c
ur
a
c
y,”
I
nf
or
m
at
io
n Sy
s
te
m
s
F
r
ont
ie
r
s
, vol
. 20, no. 3, pp. 577
–
588, Au
g. 2016, doi:
10.1007/s
10796
-
016
-
9689
-
z.
[
2]
J
. B
. R
e
be
c
c
a
, V
. V
R
a
ma
li
nga
m, V. S
uguma
r
a
n, a
nd D
. R
a
jk
u
ma
r
, “
P
r
e
di
c
ti
ve
a
na
ly
s
is
of
onl
in
e
t
e
le
vi
s
io
n vi
de
os
u
s
in
g ma
c
hi
ne
le
a
r
ni
ng
a
lg
or
it
hms
,”
F
undame
nt
al
s
and
M
e
th
ods
of
M
ac
h
in
e
and
D
e
e
p
L
e
ar
ni
ng
.
W
il
e
y,
pp.
237
–
257,
J
a
n.
2022.
doi
:
10.1002/978111982
1908.c
h10.
[
3]
Y
.
A
n,
J
.
A
n,
a
nd
S
.
C
ho,
“
A
r
ti
f
ic
ia
l
in
te
ll
ig
e
nc
e
-
ba
s
e
d
pr
e
di
c
ti
ons
of
movi
e
a
udi
e
nc
e
s
on
ope
ni
ng
S
a
tu
r
da
y,”
I
nt
e
r
nat
io
nal
J
our
nal
of
F
or
e
c
as
ti
ng
, vol
. 37, no. 1, pp. 274
–
288, J
a
n. 2021,
doi
:
10.1016/j
.i
jf
or
e
c
a
s
t.
2020.05.005.
[
4]
B
.
A
kgül
a
nd
T
.
K
üç
ükyi
lm
a
z
,
“
F
or
e
c
a
s
ti
ng
T
V
r
a
ti
ngs
of
T
ur
ki
s
h
te
le
vi
s
io
n
s
e
r
ie
s
u
s
in
g
a
two
-
le
ve
l
ma
c
hi
n
e
le
a
r
ni
ng
f
r
a
me
w
or
k,”
T
ur
k
is
h
J
ou
r
nal
of
E
le
c
t
r
ic
al
E
ngi
ne
e
r
in
g
and
C
om
put
e
r
S
c
ie
nc
e
s
,
vol
.
30,
no.
3,
pp.
750
–
766,
M
a
r
.
2022,
doi
:
10.55730/1300
-
0632.3809.
[
5]
R
.
A
kul
a
,
Z
.
W
ie
s
e
lt
hi
e
r
,
L
.
M
a
r
ti
n,
a
nd
I
.
G
a
r
ib
a
y,
“
F
or
e
c
a
s
t
in
g
th
e
s
uc
c
e
s
s
of
te
le
vi
s
io
n
s
e
r
ie
s
u
s
in
g
ma
c
hi
n
e
le
a
r
ni
ng,”
2
019
Sout
he
as
tC
on
. I
E
E
E
, A
pr
. 2019. doi:
10.1109/s
out
he
a
s
tc
on42311.2019.9020419.
[
6]
A
.
A
r
s
la
n
e
t
al
.
,
“
A
c
ompr
e
he
ns
iv
e
f
r
a
me
w
or
k
f
or
me
a
s
ur
in
g
t
he
im
me
di
a
te
im
pa
c
t
of
T
V
a
dve
r
ti
s
e
me
nt
s
:
T
V
-
I
mpa
c
t,
”
E
nt
r
opy
,
vol
. 26, no. 2, p. 109, J
a
n. 2024, doi:
10.3390/e26020109
.
[
7]
M
.
E
.
C
a
mm
a
r
a
no,
A
.
G
ua
r
in
o,
D
.
M
a
la
ndr
in
o,
a
nd
R
.
Z
a
c
c
a
gni
no,
“
T
V
s
how
s
popula
r
it
y
pr
e
di
c
ti
on
of
ge
nr
e
-
in
de
pe
nde
nt
T
V
s
e
r
ie
s
th
r
ough
ma
c
hi
ne
le
a
r
ni
ng
-
ba
s
e
d
a
ppr
oa
c
he
s
,”
M
ul
ti
m
e
di
a
T
ool
s
and
A
ppl
ic
at
io
ns
,
F
e
b.
2024,
doi
:
10.1007/s
11042
-
024
-
18518
-
z.
[
8]
E
.
C
.
A
le
má
n
C
a
r
r
e
ón,
H
.
N
ona
ka
,
A
.
H
e
nt
ona
,
a
nd
H
.
Y
a
ma
s
hi
r
o,
“
M
e
a
s
ur
in
g
th
e
in
f
lu
e
nc
e
of
me
r
e
e
xpos
ur
e
e
f
f
e
c
t
of
T
V
c
omm
e
r
c
ia
l
a
dve
r
ts
on pur
c
ha
s
e
be
ha
vi
or
ba
s
e
d on ma
c
hi
ne
l
e
a
r
ni
ng pr
e
di
c
ti
on mode
ls
,”
I
nf
or
m
at
io
n P
r
oc
e
s
s
in
g & M
anage
m
e
nt
,
vol
. 56, no. 4, pp. 1339
–
1355, J
ul
. 2019, doi:
10.1016/j
.i
pm.2019.03.007.
[
9]
C
.
E
.
G
.
M
or
e
ta
,
M
.
R
.
C
.
A
c
os
ta
,
a
nd
I
.
K
oo,
“
P
r
e
di
c
ti
on
of
di
gi
ta
l
te
r
r
e
s
tr
ia
l
te
le
vi
s
io
n
c
ove
r
a
ge
us
in
g
ma
c
hi
ne
le
a
r
ni
ng
r
e
gr
e
s
s
io
n,”
I
E
E
E
T
r
ans
ac
ti
ons
on B
r
oadc
as
ti
ng
, vol
. 65, no. 4
, pp. 702
–
712, De
c
. 2019, doi:
10.1109/t
bc
.2019.2901409.
[
10]
L
.
W
a
ng,
“
F
or
e
c
a
s
t
mode
l
of
T
V
s
how
r
a
ti
ng
ba
s
e
d
on
c
onv
ol
ut
io
na
l
ne
ur
a
l
ne
twor
k,”
C
om
pl
e
x
it
y
,
vol
.
2021,
pp.
1
–
10,
F
e
b.
2021, doi:
10.1155/2021/
6694538.
[
11]
P
.
D
a
na
he
r
a
nd
T
.
D
a
gge
r
,
“
U
s
in
g
a
ne
s
te
d
lo
gi
t
mode
l
to
f
or
e
c
a
s
t
te
le
vi
s
io
n
r
a
ti
ngs
,”
I
nt
e
r
nat
io
nal
J
our
nal
of
F
or
e
c
as
ti
ng
,
vol
. 28, no. 3, pp. 607
–
622, J
ul
. 2012, doi:
10.1016/j
.i
jf
or
e
c
a
s
t.
2012.02.008.
[
12]
Y.
-
H
.
C
he
ng,
C
.
-
M
.
W
u,
T
.
K
u,
a
nd
G
.
-
D
.
C
he
n,
“
A
pr
e
di
c
t
in
g
mode
l
of
T
V
a
udi
e
n
c
e
r
a
ti
ng
ba
s
e
d
on
th
e
F
a
c
e
book,
”
2
013
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on Soc
ia
l
C
om
put
in
g
. I
E
E
E
, S
e
p. 201
3. doi:
10.1109/s
oc
ia
lc
om.2013.167.
[
13]
T
.
T
h
a
ip
is
ut
ik
ul
a
nd
S
.
T
ua
r
ob,
“
B
e
yond
th
e
twe
e
ts
:
D
i
s
c
ove
r
in
g
f
a
c
to
r
s
th
a
t
in
f
lu
e
nc
e
T
V
s
e
r
ie
s
pr
e
f
e
r
e
nc
e
s
f
r
om
ubi
qui
to
us
s
oc
ia
l
ne
twor
ks
,”
2017
10t
h
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on
U
bi
-
m
e
di
a
C
om
put
in
g
and
W
or
k
s
hops
(
U
bi
-
M
e
di
a)
.
I
E
E
E
,
A
ug.
2
017.
doi
:
10.1109/um
e
di
a
.2017.8074106.
[
14]
C
.
A
ka
r
s
u
a
nd
B
.
D
ir
i,
“
T
ur
ki
s
h
T
V
r
a
ti
ng
pr
e
di
c
ti
on
w
it
h
T
w
it
te
r
,”
2016
24t
h
Si
gnal
P
r
oc
e
s
s
in
g
and
C
om
m
uni
c
at
io
n
A
ppl
ic
at
io
n C
onf
e
r
e
nc
e
(
SI
U
)
. I
E
E
E
, M
a
y 2016. doi:
10.1109/s
iu
.2016.7495748.
[
15]
B
.
C
i
z
m
e
c
i
a
n
d
S
.
G
.
O
gu
du
c
u,
“
P
r
e
d
ic
ti
ng
I
M
D
b
R
a
ti
n
g
s
of
P
r
e
-
r
e
l
e
a
s
e
M
ov
ie
s
w
it
h
f
a
c
to
r
iz
a
ti
o
n
m
a
c
hi
n
e
s
u
s
i
ng
s
o
c
i
a
l
me
di
a
,”
2
0
18
3
r
d
I
nt
e
r
n
at
i
on
al
C
o
nf
e
r
e
n
c
e
o
n
C
om
pu
t
e
r
S
c
ie
n
c
e
a
nd
E
ng
in
e
e
r
i
ng
(
U
B
M
K
)
.
I
E
E
E
,
20
18
.
do
i:
10
.1
10
9/
ub
mk
.2
01
8.
85
66
66
1
.
[
16]
M.
-
H
.
C
he
ng,
Y
.
-
C
.
W
u,
a
nd
M
.
-
C
.
C
he
n,
“
T
e
le
vi
s
io
n
me
e
ts
F
a
c
e
book:
T
he
c
or
r
e
la
ti
on
be
twe
e
n
T
V
r
a
ti
ngs
a
nd
s
oc
ia
l
me
d
ia
,”
A
m
e
r
ic
an J
our
nal
of
I
ndus
tr
ia
l
and B
us
in
e
s
s
M
anage
m
e
nt
, vol
.
06, no. 03, pp. 282
–
290, 2016, doi:
10.4236/ajibm
.2016.63026.
[
17]
L
.
S
ong,
Y
.
S
hi
,
a
nd
G
.
K
.
F
.
T
s
o,
“
C
omm
e
r
c
ia
l
a
udi
e
nc
e
r
e
te
nt
io
n
of
te
le
vi
s
io
n
pr
ogr
a
ms
:
me
a
s
ur
e
me
nt
a
nd
pr
e
di
c
ti
on,”
I
nt
e
r
nat
io
nal
J
our
nal
of
A
dv
e
r
ti
s
in
g
, vol
. 41, no. 3, pp. 435
–
46
1, A
pr
. 2021, doi:
10.1080/02650487.
2021.1906541.
[
18]
M
.
J
.
A
w
a
n
e
t
al
.
,
“
A
r
e
c
omm
e
nda
ti
on
e
ngi
ne
f
or
pr
e
di
c
ti
ng
movi
e
r
a
ti
ngs
us
in
g
a
bi
g
da
ta
a
ppr
oa
c
h,
”
E
le
c
tr
oni
c
s
,
vol
.
10,
no. 10, p. 1215, M
a
y 2021, doi:
10.3390/ele
c
tr
oni
c
s
10101215.
[
19]
L
.
N
ix
on,
“
P
r
e
di
c
ti
ng
your
f
ut
ur
e
a
udi
e
nc
e
’
s
popula
r
to
pi
c
s
to
opt
im
iz
e
T
V
c
ont
e
nt
ma
r
ke
ti
ng
s
uc
c
e
s
s
,”
P
r
o
c
e
e
di
ngs
of
th
e
2nd
I
nt
e
r
nat
io
nal
W
or
k
s
hop
on
A
I
fo
r
Sm
ar
t
T
V
C
ont
e
nt
P
r
oduc
ti
on,
A
c
c
e
s
s
and
D
e
li
v
e
r
y
.
A
C
M
,
O
c
t.
2020.
doi
:
10.1145/3422839.34
23062.
[
20]
S
.
T
ka
te
k,
S
.
B
a
ht
i,
Y
.
L
mz
oua
r
i,
a
nd
J
.
A
bou
c
ha
ba
ka
,
“
A
r
ti
f
ic
ia
l
in
te
ll
ig
e
nc
e
f
or
im
pr
ovi
ng
th
e
opt
im
iz
a
ti
on
of
np
-
ha
r
d
pr
obl
e
ms
:
A
r
e
vi
e
w
,”
I
nt
e
r
nat
io
nal
J
ou
r
nal
of
A
dv
anc
e
d
T
r
e
nds
in
C
om
put
e
r
Sc
ie
n
c
e
and
E
ngi
ne
e
r
in
g
,
vol
.
9,
no.
5,
pp. 7411
–
7420, Oc
t.
2020, doi:
10.30534/i
ja
tc
s
e
/2
020/
7395202
0.
[
21]
J
.
Z
hou,
Z
.
W
e
i,
B
.
P
e
ng,
a
nd
F
.
C
hi
, “
R
e
s
e
a
r
c
h a
nd
a
ppl
ic
a
ti
o
n
of
f
il
m
a
nd
te
le
vi
s
io
n
li
te
r
a
tu
r
e
r
e
c
omm
e
nda
ti
on
ba
s
e
d
on
s
e
c
ur
e
in
te
r
ne
t
of
th
in
gs
a
nd
ma
c
hi
ne
le
a
r
ni
ng,”
M
obi
le
I
nf
or
m
at
io
n
Sy
s
t
e
m
s
,
vol
.
2021,
pp.
1
–
10,
O
c
t.
2021,
doi
:
10.1155/2021/
4066267.
[
22]
P
.
Z
ha
ng,
C
.
L
iu
,
K
.
N
in
g,
W
.
Z
hu,
a
nd
Y
.
Z
ha
ng, “
P
r
e
di
c
ti
on
a
nd
pr
of
il
in
g
of
a
udi
e
nc
e
c
ompe
ti
ti
on
f
or
onl
in
e
te
le
vi
s
io
n
s
e
r
i
e
s
,
”
P
r
oc
e
e
di
ngs
of
th
e
26t
h
A
C
M
SI
G
K
D
D
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on
K
now
le
dge
D
is
c
ov
e
r
y
&
D
at
a
M
in
in
g
.
A
C
M
,
A
ug.
20
20.
doi
:
10.1145/3394486.34
03188.
[
23]
L
.
S
on
g,
Y
.
S
hi
,
G
.
K
.
F
.
T
s
o
,
a
nd
H
.
P
.
L
o,
“
F
or
e
c
a
s
ti
ng
w
e
e
k
-
to
-
w
e
e
k
t
e
l
e
vi
s
i
on
r
a
t
in
g
s
u
s
i
ng
r
e
d
uc
e
d
-
f
o
r
m
a
nd
s
t
r
u
c
t
ur
a
l
d
yn
a
m
i
c
m
od
e
ls
,
”
I
nt
e
r
n
at
io
na
l
J
o
u
r
n
al
o
f
F
o
r
e
c
a
s
ti
ng
,
vo
l.
3
7,
n
o.
1
,
pp
. 3
0
2
–
32
1,
J
a
n.
2
02
1,
d
oi
:
1
0.
10
16
/j
.i
jf
or
e
c
a
s
t.
20
20
.0
6.
00
2.
[
24]
K
.
P
r
a
de
e
p,
C
.
R
.
T
in
tu
R
os
mi
n,
S
.
S
.
D
ur
om,
a
nd
G
.
S
.
A
ni
s
ha
,
“
D
e
c
is
io
n
tr
e
e
a
lg
or
it
hms
f
or
a
c
c
ur
a
te
pr
e
di
c
ti
on
o
f
mo
vi
e
Evaluation Warning : The document was created with Spire.PDF for Python.