98
Alizadeh
In . J. Biosci.
2020
RESEARCH PAPER OPEN ACCESS
A i icial neu al ne wo ks o es ima e, a ichoke's an ioxidan
componen s e alua ion based on he easily a ailable soil
p ope ies
Azadeh Alizadeh
Facul y o Plan P oduc ion, Go gan Uni e si y o Ag icul u e and Na u al Resou ces, I an
Key wo ds: A i icial Neu al Ne wo k, A ichoke, Func ion, Medicinal plan s, An ioxidan , Soil ex u e.
h p://dx.doi.o g/10.12692/ijb/16.6.98-120
A icle published on June 16, 2020
Abs ac
One o he mos impo an equi emen s in planning p oduc ion and p ocessing o medicinal plan s in o de o
ob ain high yield and high-quali y is he ini ial assessmen o he soil physical and chemical p ope ies, which
can educe he p oduc ion cos by a oiding he use o unnecessa y soil analysis. A ichoke (Cyna a scolymus L.)
is one o he use ul and medical he bs which is conside ed as he plan quali a i e index based on he seconda y
componen s like an ioxidan componen s. The e o e, i is necessa y o e alua e he yield pe o mance o
a ichoke by means o as and cheap me hods wi h an accep able accu acy. The p esen s udy aims a
in es iga ing he amoun o an ioxidan s o a ichoke by means o soil physical and chemical cha ac e is ics
including: soil ex u e, pe cen o o ganic ca bon, pe cen o neu alizing subs ances, pH, EC, CEC, phospho us,
po assium, ni ogen and appa en speci ic g a i y by a i icial neu al ne wo k. So soil sampling conduc ed om
60 di e en ag icul u al and o es lands o Goles an P o ince, soil pa ame e s measu ed in lab. Based on
sensi i e pa ame e s di e en models ha e been designed. The esul s showed ha all a i icial neu al ne wo k
models we e mo e e icien a he han mul i a ia e eg ession model. The model 5 is selec ed wi h an o e all
iew as an op imal model, as wi h a minimum inpu pa ame e wi h a unc ion close o o he models wi h he
numbe o pa ame e s. Howe e , he numbe 4 model, because in he explana o y coe icien compa ed o he
h ee models, will be chosen, especially in he case o he pe o mance and cos o being selec ed, because wi h a
es (soil ex u e), h ee pa ame e s a e measu ed. The esul s indica ed ha he neu al ne wo k applica ion was
used o es ima e an ioxidan amoun pe o mance using soil pa ame e s, bu i is also sugges ed o con inue o
access he de ini i e esul s o simila esea ch in his ega d.
* Co esponding Au ho : Azadeh Alizadeh azadehali[email p o ec ed]
In e na ional Jou nal o Biosciences | IJB |
ISSN: 2220-6655 (P in ), 2222-5234 (Online)
h p://www.innspub.ne
Vol. 16, No. 6, p. 98-120, 2020
99
Alizadeh
In . J. Biosci.
2020
In oduc ion
A ichoke plan (Cyna a scolymus L.) is a a ie y o
composi e species. I is a pe ennial plan which g ows
o 1.5m heigh and has s aigh and g oo ed s ems.
Lea es in his plan a e big and a e di ided in o
diso de ed and jagged pieces, he su ace o he lea es
is ligh g een and unde nea h o hem is ma and
whi e colo ed due o lo s o lu s (Mahboubi, 2018).
A ichoke con ains big capi ol wi h ubula ed-pu ple
lowe s which a e enci cled a ound capi ol. The b ac s
hickened in he junc ion poin o he capi ol and
con ains nu i ion s o age. The seeds a e small wi h
g oo ed su ace and in ligh b own wi h da k b own
lines (Guadalupe e al., 2019).
A i icial neu al ne wo k is simula ed om human
neu al ne wo k and in ac is an imi a ion o human
neu al ne wo k and b ain. A emp s ha e been made
ha his ne wo k p epa es a s uc u e o be able o
lea n, gene alize and make decision like b ain (Rao
and Rao, 1996). In such s uc u es he objec i e is o
each he model and o sa e he sys em pe o mance
in model memo y by in oducing a dynamic sys em
pe o mance o use i in acing wi h cases ha en'
been me be o e. Such sys ems ha e been ecen ly
used in I an and especially in ag icul u al sciences bu
due o hei abili y o model so complica ed p ocesses
wi h many in luen ial ac o s in hem, i would be
possible o b oadly use i in ag icul u al sciences.
Among he applica ions o a i icial neu al ne wo ks
in ag icul u al sciences a e loca ion p edic ion and
p ecipi a ion ime (Kaul e al., 1999), p edic ing
ain ed whea pe o mance (Mosa a e al., 2004)
p edic ing e apo a ion and pe spi a ion (Kuma e
al., 2002) and p edic ing CO2 low in ecosys ems
(Melesse and Hanley, 2005). No s udy has been
conduc ed in he applica ion o a i icial neu al
ne wo k models in iden i ica ion o minimum inpu
pa ame e s necessa y o simula e he soil physical
and chemical p ope ies on a ichoke lea quali a i e
indices. The p esen s udy has been conduc ed on he
de e mina ion o minimum e ec i e inpu
pa ame e s on a ichoke lea quali a i e indices by
means o a i icial neu al ne wo k and he applica ion
o such ne wo ks o e alua e a ichoke quali a i e
indices by means o soil eadily a ailable pa ame e s
in Goles an zone (D ummond e al., 2003). The
esea ch objec i es a e: o de e mine he soil eadily
a ailable pa ame e s necessa y o e alua e a ichoke
pe o mance; o es ima e he e ec i eness o soil
eadily a ailable pa ame e s on a ichoke
pe o mance by a i icial neu al ne wo k; and o
de e mine he op imal model o soil eadily a ailable
pa ame e s, e ec i e on an ioxidan amoun
pe o mance.
Ma e ials and me hods
The cu en s udy conduc ed in Go gan Uni e si y o
Na u al Resou ces and Ag icul u al Sciences
G eenhouse in Oc obe 2014. The expe imen
conduc ed in andom blocks design wi h h ee
epe i ions and in lowe po s. Ag icul u al soil o 60
di e en zones a ound Goles an p o ince conside ed
as con ol which ans e ed o he lowe po s o
cul i a ion a e collec ing. Be o e beginning he
expe imen , 60 di e en zones selec ed om a ound
Go gan own o sampling, Table (1) and necessa y
ag icul u al ha es ed. Be o e cul i a ion, some o soil
ans e ed o lab o de e mine some o soil physical
and chemical p ope ies, esul s o soil decomposi ion
a e in Table (2). Soil measu ed pa ame e s we e soil
ex u e, appa en speci ic g a i y, ni ogen, phospho ,
po assium, and pe cen o o ganic ca bon, pe cen o
neu alizing subs ances (pe cen o lime), EC, pH and
CEC.
In o de o measu e o ganic ca bon, elec ical
conduc i i y and sa u a ed soil pas e acidi y me hod
ha e been used (Page e al., 1987) and hyd ome ical
me hod used o de e mine soil ex u e (Mo ahedi and
Rezaei, 1999). To measu e soil ni ogen pe cen ,
amoun s o ammonical ni ogen and ni a e ni ogen
measu ed. B emne and Mul aney (1982) me hod
used o measu e ammonical ni ogen and Page and
his colleagues (1987) me hod also used o measu e
soil po assium and sodium.
P epa ing plan samples
The lowe po s illed wi h each o collec ed soils in
h ee epe i ions. Plas ic lowe po s wi h 35cm heigh
100
Alizadeh
In . J. Biosci.
2020
and 20cm diame e used in his expe imen .
A ichoke seeds supplied om Go gan Uni e si y o
Na u al Resou ces and Ag icul u al Sciences, Facul y
o Ga dening lab. Flowe po s soils mixed wi h one
en h o lowe po olume wi h ine g ained pe li e
and hen wa e ed, a e some days 2 seeds plan ed
du ing soil la ening and co e ed wi h a hin laye o
pea moss in 2mm hickness, 7 days a e plan ing he
seeds s a ed sp ou ing. Du ing 4-lea s age hinning
conduc ed so as o keep a s ong sh ub in each
lowe po . To es ablish necessa y and uni o m
condi ions o g ow and nou ish he plan s all he
g owing ope a ion such as weeding and wa e ing
done manually, plan ing o ha es ing pe iod ook
120 days.
Measu ing mo phological cha ac e is ics and
pe o mance componen s
120 days a e plan ing, di e en con ol plan s
sampled in iden ical si ua ions. Numbe o heal hy
lea es, numbe o unheal hy lea es (yellowed, d ied
and damaged), weigh o a single sh ub, and heigh o
he sh ub and leng h o oo measu ed. To de e mine
he we weigh digi al 0.001 used and also 100cm
ule used o measu e s em and oo leng h.
Ha es ing
When a ound 98% o he lea es ma gin u ned om
smoo h and wi hou ho n in o oo hed and se a ed
s a e, hey we e ha es ed (Baghalian and Naghdi-
Badi, 2001). A e ha es ing he plan s and he
p elimina y d ying in he shadow, he lea es we e
sepa a ed and hen placed in he o en o 48 hou s in
45°C o he inal d ying and inally he b powe
p epa ed ou o hem.
Modeling h ough a i icial neu al ne wo k:
expanding an a i icial neu al ne wo k necessi a es
designing i s echnical componen s. In o de o
achie e he objec i es neu al ne wo ks wi h di e en
s uc u es like Pe cep on used o selec and apply he
bes and he mos e icien ne wo k as well as
de e mining i s e o a e. Also sensi i i y analysis
used o achie e ac o s e ec i e on an ioxidan s
pe o mance. A las he leas Random Mean Squa ed
E o (RMSE) and de e mina ion coe icien indices
used o selec he sui able and op imal model. In his
s udy soil eadily a ailable pa ame e s (soil ex u e,
pe cen o o ganic ca bon, pe cen o neu alizing
subs ances (lime pe cen ), ni ogen, phospho ,
po assium, sal iness, acidi y, ca ion exchange
capaci y, appa en speci ic g a i y) conside ed as he
inpu da a and amoun o an ioxidan s conside ed as
he ou pu da a.
Da a s anda diza ion
Basically, en e ing aw da a educes ne wo k speed
and accu acy. To a oid om such si ua ion and also
in o de o assimila e da a alues, inpu da a has o be
s anda dized be o e eaching he neu al ne wo k. This
p e en s om weigh s o become so small (Saji
Kuma and Tanda sa a, 1999) meanwhile i would be
possible o place neu ons in a desi ed ange and
p e en s om neu ons ea ly sa u a ion by egula ing
inpu da a in a speci ic ange. Also his happens o
da a u ns in o numbe s be ween 0 and 1, since mos
o h eshold unc ions’ ou pu a e numbe s be ween 0
and 1 and he o m o inpu da a o i plays an
impo an ole in ne wo k lea ning. Neu ons’ weigh
a ia ions will be he leas o inpu nea o 0 o 1, bu
neu ons’ esponse o inpu signals will be as e o
inpu amoun s nea o 0.5. The ollowing ela ionship
is used o s anda dize he da a:
Rela ionship (1)
In his ela ionship, Xn is eagen o no malized da a,
X eagen o obse a ional da a and Xmean, Xmin, Xmax
a e eagen s o mean, minimum and maximum da a
espec i ely.
Da a classi ica ion
A i icial neu al ne wo ks need a se ies o inpu and
ou pu da a o design and each o be able o ex ac
he nonlinea ela ionships h ough a ional analysis
which is done be ween hese da a as sample and do
he simula ion o simila p obable cases. A i icial
neu al ne wo ks need h ee clus e s o educa ional,
alida ion and es da a o designing. Educa ional
101
Alizadeh
In . J. Biosci.
2020
da a used o ind he ela ionship be ween
obse a ional inpu and ou pu . Valida ion da a used
o con ol and obse e he ne wo k co ec lea ning
and es da a used o e alua e he sugges ed ne wo k
e alua ion. In he cu en s udy 60%, 20% and 20%
o he o al da a assigned o eaching, alida ing and
es ing espec i ely.
Ne wo k designing
Pe cen o sand, sil and clay, o ganic ca bon pe cen ,
soil acidi y, sal iness, ni ogen, phospho , po assium,
ca ion exchange capaci y and appa en speci ic
g a i y conside ed as inpu pa ame e s an ioxidan
conside ed as he ne wo k ou pu . Then 60% o da a
(60 soil samples) selec ed o model eaching, 20% o
da a (12 soil pa ame e s) selec ed o model
alida ion p ocess and 20% (12 soil pa ame e s)
selec ed as model es da a. Teaching p ocess
including weigh a ia ion among di e en laye s
du ing eaching pe iod conduc ed o minimize he
di e ence be ween eal da a ( oe es da a) and he
p edic ed da a. The educa ional p inciple o
Le enbe g-Ma qua d and a hidden laye wi h Logsig
h eshold unc ion and Tansig h eshold unc ion o
he ou pu laye ha e been used du ing eaching
p ocess. A las he bes ne wo k s uc u es ha e been
selec ed o an ioxidan amoun based on he leas
amoun o Mean Squa ed E o (MSE) and he mos
amoun o Co ela ion Coe icien (R2).
E alua ing model accu acy
Co ela ion coe icien (R2) and RMSE used be ween
measu ed and p edic ed da a o e alua e model
pe o mance. RMSE s a is ics a e ma hema ically
explained as ollows.
Rela ionship (2)
In he abo e ela ionship, a and a e amoun s o
p edic ed and measu ed pe o mance and an ioxidan
amoun espec i ely and N is he numbe o da a.
RMSE amoun shows o wha ex en he p edic ions
ha e es ima ed he measu e mo e o less. I he
measu ed and he p edic ed amoun s be equal RMSE
will be 0. Co ela ion coe icien is also a chi ed by
line i ness be ween p edic ed da a agains measu ed
da a.
Sensi i i y analysis
Sensi i i y analysis p ocess p o ides he model
designe and a chi ec wi h aluable in o ma ion
abou model sensi i i y o inpu a iables. Iden i ying
inpu a iables e ec on model p edic ion accu acy
makes i possible o omi less e ec i e a iables om
he ne wo k and o expand and de elop a simple
model. In he o he wo ds, sensi i i y analysis is used
o de ec which o he 12 pa ame e s (inducing
pe cen o sand, sil and clay, o ganic ca bon pe cen ,
…) has had he mos e ec on pe o mance and
an ioxidan amoun and i s a ia ion has had he
mos sensi i i y. Coe icien wi hou sensi i i y
dimension used in his s udy o do model sensi i i y
analysis (Hill, 1998) as ollows: i s 12-pa ame e
model (wi hou any a ia ion in inpu ) en e ed he
ne wo k and he ou pu ex ac ed (con ol). Then one
o he a iables changed 10% and he o he ones
emained s able, he changed a iable en e ed he
ne wo k and inally he ou pu ne wo k is calcula ed.
Now he di e ence be ween hese wo ou pu s
(con ol wi h changed) is calcula ed based on
ela ionship (3-18):
Equa ion (3)
Whe e, ŷ(i+0.1) is he ou pu one o co esponding
inpu s wi h ha has changed 10% and ne wo k
p edic s i . ŷi is he ou pu ha ne wo k p edic s i
ega dless o any a ia ion in i s inpu s (con ol).
Nex , sensi i i y coe icien is calcula ed o each o
he pa e ns based on ela ionship (3-19), which is in
ac demons a i e o model sensi i i y o pa ame e
in j h obse a ional da a.
Equa ion (4)
102
Alizadeh
In . J. Biosci.
2020
Whe e, i is demons a i e o j h inpu a iable o
model.
Equa ion (5)
𝛽𝛿𝑗 is also he changed inpu which is calcula ed by
(3-20) ela ionship (in his s udy he a iables has
changed 10%). These s ages pe o m o all inpu s, in
he o he wo ds each ime o he inpu s a e changed
10% and o he a iables a e conside ed s able. In
o de o calcula e model sensi i i y, Composi e
Sensi i i y Coe icien (CSS) used o all obse a ions.
(Hill, 1998) de ined he amoun o his coe icien o
j h pa ame e as he ollowing;
Equa ion (6)
Equa ion (3-22) is in ac he sensi i i y coe icien
a e age o each inpu . Fo simple compa ison,
di e en a iables CSS amoun s has used CSS ela i e
amoun known as ela i e sensi i i y coe icien (γ) as
he ollowing:
Equa ion (7)
Whe e, max (CSS) is he maximum amoun o CSS o
all inpu a iables o he model. The mos amoun is
equal wi h he uni and is ela ed o a pa ame e wi h
he maximum o CSS.
Da a analysis
SPSS employed o compa e di e en con ol g oup
cha ac e is ics mean. Also in o de o e alua e
pe o mance (weigh o single sh ub) o a ichoke,
Ma lab a i icial neu al; ne wo k so wa e was used.
Resul s and discussion
Va iables desc ip ion: expe imen al and ield da a
ha e o be o de ed as a mass o aw numbe s o each
ype o s a is ical s udy o calcula ion.
Regula ing nume ical da a and d awing hei diag am
is he i s s age o s a is ical analysis.
Table 1. S a is ical desc ip ion o soils’ chemical p ope ies.
Pa ame e
Minimum
Maximum
Mean
Coe icien o a ia ion
Chologi
O ganic ca bon
0.12
3.97
1.21
0.55
2.08
TNV %
0.5
36.3
15
0.69
0.26
CEC
7.02
26.2
10.85
0.26
2.80
EC
0.3
1.88
0.47
0.47
4.95
pH
6.12
8.08
7.67
0.038
3.59
N
0.02
0.9
0.12
0.89
5.64
P
4
86
16.7
1.04
2.94
K
106
754
303.47
0.40
1.07
This da a con ains impo an and use ul in o ma ion
un il be o de ed. Tables 1 and 2 ha e summa ized he
s a is ical desc ip ion o soils physical and chemical
p ope ies ela ed o 60 soil samples, espec i ely.
Soil a iabili y is he key elemen in soil spa ial
speci ic managemen and p o ides aluable
in o ma ion abou he na u e o soil cha ac e is ics in
a ms (Ayoubi e al., 2008). Acco ding o Table 1, pH
has he leas a ia ions coe icien (0.038%) among
physical and chemical a iables. Howe e , P
a ia ions coe icien has been highe and equal wi h
1.04 among chemical a iables.
Skewness coe icien amoun s in Tables 1 and 2
demons a es ha besides lime, po assium, sand, sil ,
clay and BD pa ame e s which ha e no mal
dis ibu ion and ha e -1 o +1 skewness coe icien ,
o he pa ame e s ha e no mal log dis ibu ion.
103
Alizadeh
In . J. Biosci.
2020
A i icial neu al ne wo k modeling
This sec ion shows esul s o he bes a i icial neu al
ne wo k s uc u e by 12 inpu pa ame e s o 60 soil
samples. Also sensi i i y analysis and i s esul s a e
ep esen ed in he ollowing. This s udy has used
mul ilaye pe cep on and ans e unc ion in hidden
laye and ou pu laye , numbe o hidden laye s,
numbe o neu ons in hidden laye o ne wo k we e
expe imen and he bes s uc u e o an ioxidan
amoun achie ed by ial and e o .
Selec ing he bes ne wo k o p edic ing an ioxidan
amoun is also based on he leas amoun o RMSE
and he mos amoun o R2.
Table 2. S a is ical desc ip ion o soils' physical p ope ies.
Pa ame e
Minimum
Maximum
Mean
Coe icien o Va ia ion (CV)
Chologi
Sand
12
74
32.05
0.43
0.55
Sil
11
83
44.27
0.31
0.17
Clay
4
39
23.23
0.38
-0.35
Bd
0.85
2.16
1.66
0.16
-1.38
The bes a angemen o hidden laye wi h
Le enbe g-Ma qua d educa ional loga i hm as a
hidden laye , 34 neu ons, Logsig h eshold unc ion
o hidden laye and Tansig o ou pu laye ha e
been selec ed in modeling an ioxidan amoun wi h 12
pa ame e s o 60 soil samples. Table 3 shows
calcula ed s a is ical pa ame e s o gene a ed model
du ing eaching, alida ion, es ing and o al s ages
o an ioxidan amoun espec i ely.
I is possible o ob ain in o ma ion abou model
pe o mance h ough gi en i ness line g adien
amoun be ween p edic ed da a agains measu ed
da a.
Table 3. Calcula ed s a is ical pa ame e s o s ages o eaching, alida ion, es ing and o al in 12-pa ame e
model o an ioxidan componen s.
Le el
R2
RMSE
Educa ion
0.99
0.001
Valida ion
0.99
0.003
Tes
0.99
0.003
To al
0.99
0.002
Table 4. Resul s o soil eadily a ailable sensi i i y analysis o an ioxidan componen s.
Pa ame e
Rela i e sensi i i y coe icien (γ)
pH
0.66
OC (%)
0.89
K
0.64
TNV (%)
0.96
CEC
0.58
EC
0.48
Bd
0.47
N
0.52
Clay
0.59
Sil
1
Sand
0.61
P
0.60
104
Alizadeh
In . J. Biosci.
2020
I gi en i ness line g adien be 1 be ween p edic ed
da a agains measu ed da a, i demons a es ha
p edic ed amoun s a e equal wi h he measu ed
amoun s. Fig. 1 show gi en i ness line equa ion
be ween p edic ed da a wi h model agains measu ed
da a o an ioxidan amoun du ing s ages o model
eaching and es ing. Acco ding o Fig. 1 gi en i ness
line g adien is 1 and 1 o an ioxidan amoun
espec i ely which demons a es he app oxima ion
o p edic ed da a wi h measu ed ones and since R2 is
0.99 and 0.99 in eaching and es ing s ages
espec i ely, i is concluded ha acco ding o high
amoun o R2 and less amoun o RMSE his model's
e alua ion enjoys good accu acy.
Table 5. Design o di e en a i icial neu al ne wo k models due o he sensi i i y analysis and minimum
numbe o es s o ob ain model inpu s.
Model
Inpu pa ame e
Numbe o expe imen s
Model 1
O ganic ca bon
1
Model 2
TNV%
1
Model 3
PH
1
Model 4
Tex u e
1
Model 5
O ganic ca bon + lime %
2
Model 6
PH + O ganic ca bon
2
Model 7
PH + lime %
2
Model 8
Tex u e + O ganic ca bon
2
Model 9
PH + O ganic ca bon + lime %
3
Model 10
PH + O ganic ca bon + lime %
4
Table 6. S a is ical pa ame e s calcula ed o he s ages o educa ion, alida ion, es and o al in he model (1)
o an ioxidan componen s.
Le el
R2
RMSE
Educa ion
0.20
0.078
Valida ion
0.17
0.086
Tes
0.15
0.073
To al
0.18
0.079
In his s age, a e y accu a e model has achie ed o
e alua e amoun o an ioxidan by means o 12 eadily
a ailable pa ame e s o soil. Bu since he objec i e o
he cu en s udy is he as and easy e alua ion (less
ime and cos ) o an ioxidan amoun by means o soil
eadily a ailable pa ame e s, so applying his model
con adic s wi h he objec i es o he p esen s udy,
because i akes so much ime and cos o measu e 12
pa ame e s. The e o e, i is possible o iden i y
pa ame e s sensi i e o he e alua ion o an ioxidan
amoun by means o sensi i i y analysis and hen
di e en models whose inpu s a e gene a ed by
means o he leas numbe o expe imen s and less
pa ame e s by means o hese sensi i e pa ame e s as
he model inpu s in o de o e alua e an ioxidan
amoun pe o mance.
Table 7. S a is ical pa ame e s calcula ed o he s ages o educa ion, alida ion, es and o al in Model (2) o
an ioxidan componen s.
Le el
R2
RMSE
Educa ion
0.56
0.064
Valida ion
0.28
0.089
Tes
0.22
0.092
To al
0.41
0.076
105
Alizadeh
In . J. Biosci.
2020
Sensi i i y analysis
In he expe imen s ela ed o plan , chlo ophyll,
ca o enoid, an ioxidan , phenol, la onoid and
an ioxidan amoun measu ed, bu due o he
impo ance o an ioxidan amoun (pe o mance)
only he ow pa ame e s modeled. Acco ding,
sal iness was he mos sensi i e pa ame e o
chlo ophyll, pe cen o neu alizing subs ances (lime
pe cen ) o ca o enoid, o ganic ca bon pe cen o
phenol and acidi y o la onoid.
Table 8. S a is ical pa ame e s calcula ed o he s ages o educa ion, alida ion, es and o al in Model (3) o
an ioxidan componen s.
Le el
R2
RMSE
Educa ion
0.18
0.082
Valida ion
0.15
0.086
Tes
0.10
0.075
To al
0.15
0.082
Table 9. S a is ical pa ame e s calcula ed o he s ages o educa ion, alida ion, es and o al in Model (4) o
an ioxidan componen s.
Le el
R2
RMSE
Educa ion
0.79
0.047
Valida ion
0.51
0.080
Tes
0.37
0.090
To al
0.63
0.065
A e modeling pe o mance amoun o an ioxidan
wi h 12 pa ame e s by a i icial neu al ne wo k and
achie ing he bes ne wo k ega ding s a is ical
pa ame e s, sensi i i y analysis wi hou sensi i i y
dimension (Hill, 1998) conduc ed. Tables 4 show
sensi i i y esul s o single sh ub weigh
pe o mance. Hill (1988) main ained ha i a
pa ame ic sensi i i y coe icien be mo e han 0.1,
ha pa ame e belongs o model sensi i e
pa ame e s. Acco ding o Hill (1998), pe o mance
an ioxidan amoun is sensi i e o ail pa ame e s.
To al sensi i i y coe icien s o soil 12 di e en
pa ame e s has o be he basis o modeling o bo h
ou pu pa ame e s o pe o mance an ioxidan
amoun . Nu since he objec i e o his s udy is o
es ima e an ioxidan amoun wi h he leas numbe o
expe imen s and he necessa y pa ame e s, Table 4
de e mined he mos sensi i e pa ame e ega ding
p io i y in e alua ing an ioxidan amoun and
acco dingly e alua ing an ioxidan amoun ha e been
modeled. Based on Table 4 pa ame e s coe icien
amoun s has dec eased (pa ame e s sensi i i y has
been dec eased).
The esul s show ha a ichoke’s pe o mance
an ioxidan amoun has he mos sensi i i y o pH
and has he leas sensi i i y o soil phospho and
a ichoke’s o soil appa en speci ic g a i y.
Table 10. S a is ical pa ame e s calcula ed o he s ages o educa ion, alida ion, es and o al in Model (5) o
an ioxidan componen s.
Le el
R2
RMSE
Educa ion
0.89
0.038
Valida ion
0.83
0.049
Tes
0.80
0.053
To al
0.86
0.044
106
Alizadeh
In . J. Biosci.
2020
Designing di e en models o a i icial neu al
ne wo k by sensi i e pa ame e s
As men ioned ea lie in his s udy, he objec i e o
his s udy was o e alua e less, mo e a ailable and
na u ally lowe cos inpu pa ame e s, pe o mance
an ioxidan amoun . Thus, Table 5 shows how 4 soil
eadily a ailable pa ame e s which ha e mo e
sensi i i y in e alua ing an ioxidan amoun ha e
gene a ed di e en models o a i icial neu al
ne wo ks by inc easing he numbe o inpu
pa ame e s and he numbe o conduc ed expe imen
espec i ely.
Table 11. S a is ical pa ame e s calcula ed o he s ages o educa ion, alida ion, es and o al in model (6) o
an ioxidan componen s.
Le el
R2
RMSE
Educa ion
0.66
0.058
Valida ion
0.52
0.073
Tes
0.50
0.076
To al
0.59
0.065
Table 12. S a is ical pa ame e s calcula ed o he s ages o educa ion, alida ion, es and o al in model (7) o
an ioxidan componen s.
Le el
R2
RMSE
Educa ion
0.84
0.043
Valida ion
0.72
0.058
Tes
0.61
0.056
To al
0.78
0.049
In he model (1) due o he maximum sensi i i y
coe icien o o ganic ca bon, only he o ganic ca bon
pa ame e is c ea ed as he a i icial neu al ne wo k
model.
In he model (2), wi h he Lime pe cen pa ame e ,
he a i icial neu al ne wo k model is c ea ed.
In he model (3) wi h he pH pa ame e , he a i icial
neu al ne wo k model is c ea ed.
In model (4) wi h he pa ame e s o clay, sil and
g a el, he a i icial neu al ne wo k model is c ea ed.
In model (5) he lime pe cen age pa ame e is added
o he model (1).
In he model (6), he pH pa ame e was added o he
model (1).
In he model (7), he pH pa ame e (2) was added o
he model.
In model (8) he ex u e pa ame e was added o he
model (1).
In he model (9), he pH pa ame e was added o he
model (5).
In models (10), he ex u e pa ame e was added o
he model (9).
Table 13. S a is ical pa ame e s calcula ed o he s ages o educa ion, alida ion, es and o al in model (8) o
an ioxidan componen s.
Le el
R2
RMSE
Educa ion
0.99
0.002
Valida ion
0.83
0.056
Tes
0.69
0.078
To al
0.88
0.043
113
Alizadeh
In . J. Biosci.
2020
Fig. 11. The i ed line be ween he p edic ed da a in on o he da a measu ed by he wo emale ope a ions in
he es s age in he model (5).
Fig. 12. The i ing line be ween he p edic ed da a in he on o he measu ed da a o an ioxidan componen s
du ing he aining s age in model (6).
The slope a e o he line in his model has no
di e ence in he s ages o aining and es in bo h
an ioxidan amoun compa ed o o he models.
Acco ding o he men ioned poin s, he le el o
explana ion in an ioxidan amoun pe o mance was
be e han he mul i a ia e eg ession esul s a he
es s age.
Model esul s (6): In model numbe (6) he
an ioxidan amoun was es ima ed acco ding o he
pa ame e s o he o ganic pH and ca bon. The bes
hidden laye makeup wi h Ma kwa -Le enbe g
educa ional algo i hm was selec ed as a hidden laye ,
34 neu on, LOGSIG h eshold unc ion o hidden
laye and Tansig o he ou pu laye . Table 11
indica es s a is ical pa ame e s calcula ed o he
s ages o aining, alida ion, es ing and o al in
model (6) o an ioxidan amoun pe o mance.
Figu es 12 and 13 show R2 index and he equa ion o
he i ed line be ween he p edic ed da a agains he
measu ed da a o an ioxidan amoun in he s ages o
aining and es o model (6). As seen in he Fig. 12
and 13, he slope o he i ed line o an ioxidan
amoun pe o mance is 0.97 and 1, espec i ely. Also,
he explana o y coe icien in aining and es ing
114
Alizadeh
In . J. Biosci.
2020
s ages in an ioxidan amoun pe o mance we e 0.66
and 0.50, espec i ely. Acco ding o he men ioned
poin s, his model does no ha e accep able accu acy
o es ima ing an ioxidan amoun pe o mance. Also,
he le el o explana ion in an ioxidan amoun
pe o mance was be e han he mul i a ia e
eg ession esul s a he es s age.
Fig. 13. The i ing line be ween he p edic ed da a in on o he measu ed da a o an ioxidan componen s in
he es s age in he Model (6).
Fig. 14. The i ed line be ween he p edic ed da a in on o he measu ed da a o he wo emale ope a ions in
he aining s age in he model (7).
Model esul s (7): In model numbe (7) he
an ioxidan amoun was es ima ed based on he lime
and pH pa ame e s. The bes hidden laye makeup
wi h Ma kwa -Le enbe g educa ional algo i hm was
selec ed as a hidden laye , 45 neu on, LOGSIG
h eshold unc ion o hidden laye and Tansig o he
ou pu laye . Table 12 shows s a is ical pa ame e s
calcula ed o he s ages o educa ion, alida ion,
es ing and o al in model (7) o an ioxidan amoun
pe o mance. Fig. 14 and 15 show R2 index and he
equa ion o he i ed line be ween he p edic ed da a
agains he measu ed da a o an ioxidan amoun in
he s ages o aining and es o model (7). As shown
in he Fig. 14 and 15, he slope o he i ed line o
an ioxidan amoun pe o mance is 0.98 and 1. Also,
he explana o y coe icien in aining and es ing
115
Alizadeh
In . J. Biosci.
2020
s ages in an ioxidan amoun pe o mance we e 0.84
and 0.61, espec i ely. Bu his model is no
accep able o an ioxidan amoun . Also, he le el o
explana ion in an ioxidan amoun was be e han
he mul i a ia e eg ession esul s a he es s age.
Model esul s (8): In model numbe (8) he amoun
o an ioxidan amoun was es ima ed based on soil
and o ganic ca bon ex u e pa ame e s.
Fig. 15. The i ing line be ween he p edic ed da a in on o he measu ed da a o an ioxidan componen s in
he es s age in he model (7).
Fig. 16. The i ing line be ween he p edic ed da a in on o he measu ed da a o an ioxidan componen s
du ing he aining s age in model (8).
The bes hidden laye makeup wi h Ma kwa -
Le enbe g educa ional algo i hm was selec ed as a
hidden laye , 34 neu on, LOGSIG h eshold unc ion
o hidden laye and Tansig o he ou pu laye . Table
13 indica es s a is ical pa ame e s calcula ed o he
s ages o aining, alida ion, es ing and o al in
model (8) o an ioxidan amoun pe o mance. Fig.
16 and 17 indica e R2 index and he equa ion o he
i ed line be ween he p edic ed da a agains he
measu ed da a o an ioxidan amoun in he aining
and es s ages o model (8). As shown in Fig. 16 and
17, he slope o he i ed line o an ioxidan amoun
pe o mance is 0.99 and 1.01. Also, he explana o y
coe icien in aining and es ing s ages in
an ioxidan amoun pe o mance we e 0.99 and 0.69
le el o explana ion in an ioxidan amoun
116
Alizadeh
In . J. Biosci.
2020
pe o mance was be e han he mul i a ia e
eg ession esul s a he es s age. The esul s o
a i icial neu al ne wo k models wi h h ee
expe imen s conduc ed in ob aining he model
inpu s: Model Resul s (9): in model numbe (9) he
amoun o an ioxidan amoun was es ima ed based
on he pa ame e s o o ganic ca bon, he pe cen age
o lime and pH.
Fig. 17. The i ing line be ween he p edic ed da a in on o he measu ed da a o an ioxidan componen s in
he es s age in he model (8).
Fig. 18. The i ing line be ween he p edic ed da a in on o he measu ed da a an ioxidan componen s du ing
he aining s age in model (9).
The bes hidden laye makeup wi h Ma kwa -
Le enbe g educa ional algo i hm was selec ed as a
hidden laye , 34 neu on, LOGSIG h eshold unc ion
o hidden laye and Tansig o he ou pu laye . Table
14 indica es s a is ical pa ame e s calcula ed o he
s ages o aining, alida ion, es ing and o al in
model (9) o an ioxidan amoun pe o mance. Fig.
18 and 19 indica e R2 index and he balanced line
equa ion be ween he p edic ed da a agains he
measu ed da a o an ioxidan amoun in he s ages o
aining and es o model (9). As seen in Fig. 18 and
19, he slope o he i ed line o he ope a ion o he
an ioxidan amoun is 0.99 and 1.01. Also, he
explana o y coe icien in aining and es ing s ages
117
Alizadeh
In . J. Biosci.
2020
in an ioxidan amoun pe o mance we e 0.86 and
0.79, espec i ely. This model was accep able o
an ioxidan amoun . Also, he le el o explana ion in
an ioxidan amoun pe o mance was be e han he
mul i a ia e eg ession esul s a he es s age.
A i icial neu al ne wo k models wi h ou
expe imen s conduc ed in ob aining he model inpu s
Model esul s (10): In model numbe (10) he amoun
o an ioxidan amoun was es ima ed based on he
pa ame e s o o ganic ca bon, soil ex u e, limes one
pe cen age and pH. The bes hidden laye makeup
wi h Ma kwa -Le enbe g educa ional algo i hm was
selec ed as a hidden laye , 34 neu on, LOGSIG
h eshold unc ion o hidden laye and Tansig o he
ou pu laye .
Fig. 19. The i ing line be ween he p edic ed da a in on o he measu ed da a o an ioxidan componen s in
he es s age in he model (9).
Fig. 20. The i ing line be ween he p edic ed da a in he on o he measu ed da a o an ioxidan componen s
du ing he aining s age in model (10).
Table 15 indica es s a is ical pa ame e s calcula ed o
he s ages o educa ion, alida ion, es ing and o al in
Model (10) o an ioxidan amoun pe o mance. Fig.
20 and 21 show R2 index and he i -line equa ion o
he p e-p ojec ed da a we e measu ed in on o he
measu e o plan weigh pe o mance in aining and
es s ages o model (10). As shown in he Fig. 20 and
21, he slope o he i ed line o an ioxidan amoun
118
Alizadeh
In . J. Biosci.
2020
pe o mance, espec i ely 1 and 0.98 espec i ely,
ha exp ession is app oaching he p edic ed alues
wi h he measu ed da a in he model (10). Also, he
explana o y coe icien in aining and es ing s ages
in an ioxidan amoun pe o mance we e 0.99 and
0.78, espec i ely. This model has an accep able
esul o an ioxidan amoun . Also, he le el o
explana ion in an ioxidan amoun pe o mance was
be e han he mul i a ia e eg ession esul s a he
es s age.
Fig. 21. The i ing line be ween he p edic ed da a in on o he measu ed da a o an ioxidan componen s in
he es s age in he model (10).
Compa ison o he esul s o models designed wi h
sensi i e pa ame e s
As men ioned in he p e ious sec ions, he pu pose o
his s udy was o c ea e a ne wo k wi h he leas
numbe o es s ( he minimum numbe o ea ly
pa ame e s ound) o in oduce he bes model o
pe o mance es ima es an ioxidan amoun . Fo his
pu pose, by compa ing he esul s o 10 models
designed as well as mul i a ia e eg ession model, he
bes model o an ioxidan amoun pe o mance
es ima es a e de e mined.
Acco ding o he Table 16, wi h a gene al iew o he
addi ion o inpu pa ame e s, inc ease he amoun o
R2 and educe he amoun o RMSE in he s ages o
aining, alida ion and es ing on an ioxidan
amoun pe o mance, indica ing he imp o emen o
he model accu acy, by inc easing he numbe o
inpu pa ame e s in he an ioxidan amoun
es ima ion, i is ha is qui e ob ious.
This esul is also seen in esea ches o Shop e al.,
(1998), Shop and he Lich (1998), Moazen Zadeh e
al., (1388). All neu al ne wo k models o es ima e
an ioxidan amoun pe o mance we e be e
compa ed wi h mul i a ia e eg ession model 1-4
models ha e a nea ly simila unc ion. Howe e , he
model 5 is selec ed wi h an o e all iew as an op imal
model, as wi h a minimum inpu pa ame e wi h a
unc ion close o o he models wi h he numbe o
pa ame e s. Howe e , he numbe 4 model, because
in he explana o y coe icien compa ed o he h ee
models, will be chosen, especially in he case o he
pe o mance and cos o being selec ed, because wi h
a es (soil ex u e), h ee pa ame e s a e measu ed.
The esul s indica ed ha he neu al ne wo k
applica ion was used o es ima e an ioxidan amoun
pe o mance using soil pa ame e s, bu i is also
sugges ed o con inue o access he de ini i e esul s
o simila esea ch in his ega d.
Conclusion
The esul s showed ha he me hod o a i icial
neu al ne wo k has high accu acy in es ima ing
an ioxidan componen s A ichoke, so ha in se en
models o 10 models (explaining coe icien in he es
119
Alizadeh
In . J. Biosci.
2020
s age), change he an ioxidan componen s in he
s udied a ea using 12 cha ac e is ics. An ioxidan
unc ion depends la gely on acidi y, o ganic ca bon,
po assium and soil lime pe cen age.
This s udy showed ha he acidi y pa ame e o he
o de is he mos impo an ac o a ec ing
an ioxidan componen s pe o mance in he egion.
Also, he pe cen age pa ame e o sil was iden i ied
as he mos e ec i e ac o in an ioxidan unc ion.
The esul s ob ained in his s udy a e only a ailable
o he s udied a ea and o he simila a eas in e ms
o opog aphy, clima e, soil and manage ial
ope a ions. Howe e , i can be done like such a s udy
using a i icial neu al ne wo ks in o he a eas.
Re e ences
Ayoubi S, Kho mali F, Sah awa K. 2009.
Rela ion o ba ley biomass and g ain yields o soil
p ope ies wi hin a ield in he a id egion: Use o
ac o analysis. Ac a Ag icul u ae Scandina ica.
59(2), 107-117.
h ps://doi.o g/10.1080/0906471080193.2417
B emne JS, Mul aney CS. 1982. Ni ogen- o al.
In A. L. Page (Ed.), Me hods o Soil Analysis, Pa 2.
Ame ican Socie y o Ag onomy (p. 595-624).
Madison, Wisconsin.
h p://dx.doi.o g/10.13031/2013.12541.
D ummond ST, Suddu h KA, Joshi A, Bi ell
SL, Ki chen NR. 2003. S a is ical and Neu al
Me hods o Si e-speci ic Yield P edic ion.
T ansac ions o he Ame ican Socie y o Ag icul u al
Enginee ing 46(1), 5-14.
h p://dx.doi.o g/10.13031/201312541
Guadalupe M, Susana N, Diéguez Ma ía B,
Fe nández Paggi Ma ía B, Riccio Denisa S,
Pé ez G, Edga do R, Fabián A, Aman o Ma ía
O, Tapia Alejand o L. 2019. E ec o
os omycin, Cyna a scolymus ex ac , deoxyni alenol
and hei combina ions on in es inal heal h o weaned
pigle s. Animal Nu i ion 5(4), 386-395.
h p://dx.doi.o g/10.1016/j.aninu.2019.08.001
Hill M. 1998. Me hods and guidelines o e ec i e
model calib a ion. U.S. Geological su ey Wa e -
Resou ces In es iga ions Rep. 98-4005.
Kaul M, Hill RL, Wal hall C. 2005. A i icial
neu al ne wo ks o co n and soybean yield
p edic ion. Ag icul u e Sys ems 85, 1-18.
h p://dx.doi.o g/10.1016/j.agsy.2004.07.009
Kuma N, Raghuwanshi NS, Singh R,
Wallende WW, P ui WO. 2002. Es ima ing
e apo anspi a ion using a i icial ne wo k. Jou nal
o I iga ion and D ainage Enginee ing. 128(4),
224-233.
h p://dx.doi.o g/10.1061/(ASCE)07339437(2007)13
32(83)
Melesse AM, Hanley RS. 2005. a i icial neu al
ne wo k applica ion o mul i-ecosys em ca bon lux
simula ion. Ecological Modeling 189, 305-314.
h p://dx.doi.o g/10.1016/j.ecolmodel.2005.03.014
Moazenzadeh R, Ghah aman B, Fa halian F,
Khoshnoodiyazdi A. 2009. E ec o ype and
numbe o inpu a iables on es ima ion o mois u e
e en ion cu e and sa u a ed hyd aulic conduc i i y.
Jou nal o Wa e and Soil 23(3), 57-70.
h p://dx.doi.o g/10.1016/j.jhyd ol.2019.05058
Mahboubi M. 2018. Cyna a scolymus (a ichoke)
and i s e icacy in managemen o obesi y. Bulle in o
Facul y o Pha macy, Cai o Uni e si y 56(2), 115-
120.
h p://dx.doi.o g/10.1016/j.b opcu.2018.10.003
Mo ahedi Naeini SA, Rezaei M. 2008. Soil
Physics (Fundamen als and Applica ions). Go gan
Uni e si y o Ag icul u al Sciences and Na u al
Resou ces Publica ions, 474 p.
Musa ah B, Khalili A. 2003. P edic ion o ain ed
whea yield using a i icial neu al ne wo ks. Ni a .
48, 47-62.
h p://dx.doi.o g/10.1080/09064710903005.682
120
Alizadeh
In . J. Biosci.
2020
Page A, Mille R, Keeney D. 1982. Me hods o Soil
Analysis.2 h ed. Pa 2: Chemical and biological
p ope ies. Soil Science Socie y o Ame ica, 24 p.
Rao V, Rao H. 1996. C++ Neu al ne wo ks and
uzzy logic, BPB, New Dehli, India. 380-381.
Schaap M, Leij F. 1998. Using neu al ne wo ks o
p edic soil wa e e en ion and soil hyd aulic
conduc i i y. Soil and Tillage Resea ch 47, 37-42.
h ps://doi.o g/10.1016/S0167-1987(98)00070-1
Schaap M, Leij F, Van Genuch en M. 1998.
Neu al ne wo k analysis o hie a chical p edic ion o
soil hyd aulic p ope ies. Soil Science Socie y o
Ame ica Jou nal 62, 847-855.
h p://dx.doi.o g/10.2136/sssaj1998.0361599500620
0040001x
