Jou nal o Theo e ical and Applied In o ma ion Technology
31s May 2025. Vol.103. No.10
© Li le Lion Scien i ic
ISSN: 1992-8645 www.ja i .o g E-ISSN: 1817-3195
4191
SELF-RELIANT RESIDUAL NETWORK BASED DEEP
LEARNING FRAMEWORK FOR MELANOMA SKIN
DISEASE DETECTION
V. RADHIKA1, A. MUTHUCHUDAR2, M. LINGARAJ3
1P incipal, 2 Assis an P o esso , Associa e P o es so Depa men o Compu e Science,
Sanka a College o Science and Comme ce, India
E-mail: 1 adhika [email protected], 2mu huchuda a@sanka a.ac.in, 3linga ajm@sanka a.ac.in
ABSTRACT
Melanoma is one o he deadlies ypes o skin cance and one o he mos agg essi e ha may be i caugh
la e. While adi ional app oaches may ha e hei limi s, an accu a e diagnosis is i al o pa ien su i al.
Thanks o i s capaci y o unde s and in ica e pa e ns om massi e da ase s, deep lea ning has e ol ed as a
po en ial me hod o au oma ed melanoma diagnosis. The challenges pe sis , including o e i ing,
ins abili y du ing aining, and di icul ies in handling nonlinea i ies, which can hinde accu a e p edic ions.
To add ess hese challenges, Sel -Relian ResNe (SR-ResNe ) has been p oposed. This enhanced e sion
o ResNe in eg a es Zou endijk’s Me hod, a nonlinea op imiza ion echnique, o op imize weigh upda es
and imp o e con e gence. SR-ResNe ea u es a se ies o esidual blocks whe e Zou endijk’s Me hod
e ines he lea ning p ocess, ensu ing s abili y and e icien aining, e en in deepe ne wo ks. The
ne wo k’s a chi ec u e has been designed o enhance pe o mance and gene aliza ion. The p oposed SR-
ResNe has been e alua ed using a da ase o 10,000 Melanoma Skin Cance images. The esul s
demons a e signi ican imp o emen s in classi ica ion accu acy, achie ing a high p ecision a e wi h
educed o e i ing. SR-ResNe ou pe o ms adi ional models, es ablishing i sel as a obus ool o
melanoma diagnosis.
Keywo ds: Melanoma Skin Cance , Deep Lea ning, SR-ResNe , Zou endijk’s Me hod, Classi ica ion
Accu acy
1. INTRODUCTION
Melanoma Skin Cance has eme ged as a
c i ical public heal h conce n, ecognized o i s
agg essi e na u e and high mo ali y a e i no
diagnosed ea ly. This cance , o igina ing in
melanocy es— he cells esponsible o p oducing
melanin—demands ea ly de ec ion and p ecise
diagnosis o imp o e su i al a es[1]. T adi ional
me hods, elying on isual examina ion and biopsy,
ha e exhibi ed limi a ions, po en ially leading o
delays in ea men . The onse o he COVID-19
pandemic has u he exace ba ed his issue,
dis up ing ou ine heal hca e se ices and causing a
g adual inc ease in Melanoma Skin Cance cases as
pa ien s ha e missed ea ly sc eenings. This su ge
has unde sco ed an u gen need o mo e eliable
and ad anced diagnos ic ools ha enhance
accu acy and accessibili y in he pos -pandemic
e a[2], [3].
Figu e 1a shows a benign skin lesion, such
as a mole o benign ne us, which is non-cance ous
and does no e ol e in o melanoma. These lesions
ypically do no in ade su ounding issues o
sp ead o o he pa s o he body, making hem
gene ally ha mless.Figu e 1b, on he o he hand,
depic s a malignan skin lesion, speci ically
melanoma, a dange ous o m o skin cance [4].
Melanoma can g ow agg essi ely, in ade nea by
issues, and me as asize o dis an o gans. Accu a e
diagnosis and ea ly in e en ion a e c ucial o
e ec i e ea men o melanoma[4].
Deep lea ning has gained p ominence as a
ans o ma i e app oach in medical imaging,
e olu ionising how melanoma is de ec ed and
classi ied. Con olu ional Neu al Ne wo ks (CNNs)
ha e eme ged as a leading deep lea ning
a chi ec u e, showing signi ican p omise in
analyzing complex pa e ns wi hin medical
images[5]. Despi e hese imp o emen s, melanoma
de ec ion using deep lea ning models has been qui e
di icul . One o he bigges p oblems is o e i ing,
which occu s when a model does e y well on
aining da a bu doesn' adap well o new,
unknown da a. A lack o s abili y in he aining
p ocess has also been a signi ican hu dle, o en
Jou nal o Theo e ical and Applied In o ma ion Technology
31s May 2025. Vol.103. No.10
© Li le Lion Scien i ic
ISSN: 1992-8645 www.ja i .o g E-ISSN: 1817-3195
4192
esul ing in di icul ies wi h con e gence in deepe
ne wo ks. [6], [7].
Figu e 1a. Benign Skin Lesion
Figu e 1b. Malignan Skin Lesion
Sel -Relian Residual Ne wo k (SR-
ResNe ) has been p oposed o o e come hese
challenges, ep esen ing an ad ancemen o e he
adi ional ResNe a chi ec u e. SR-ResNe
in eg a es Zou endijk’s Me hod, a nonlinea
op imiza ion echnique designed o enhance he
aining p ocess. While ResNe is known o i s
powe ul a chi ec u e, which includes esidual
connec ions ha mi iga e he anishing g adien
p oblem in deep ne wo ks, SR-ResNe builds upon
his by op imizing weigh upda es o imp o e
con e gence and s abili y. In eg a ingZou endijk’s
Me hod ensu es ha each aining s ep mo es he
model owa d an op imal solu ion, e ec i ely
add essing issues ela ed o o e i ing and
ins abili y ha ha e p e iously hinde ed deep-
lea ning models in melanoma de ec ion.
By e ining he lea ning p ocess, SR-
ResNe has es ablished i sel as a obus and
eliable ool o melanoma de ec ion, o e ing
signi ican imp o emen s in accu acy and s abili y.
This model’s abili y o add ess he speci ic
challenges posed by melanoma da a posi ions i as a
c i ical ad ancemen in he ield, pa icula ly du ing
he COVID-19 pandemic, whe e he need o ea ly
and accu a e cance de ec ion has become
inc easingly p essing. The heal hca e landscape
con inues o e ol e in esponse o ongoing
challenges, wi h SR-ResNe p o iding a p omising
solu ion o imp o ing pa ien ou comes in
melanoma diagnosis, helping o b idge he gap le
by adi ional me hods and ensu ing ha mo e
pa ien s ecei e imely and accu a e ca e.
Despi e ad ancemen s in melanoma
de ec ion, exis ing models su e om o e i ing,
uns able con e gence, and limi ed gene aliza ion,
especially in deepe ne wo ks. Ensemble and
ans e lea ning me hods o e imp o emen s bu
lack op imiza ion e iciency and adap abili y.
Cu en app oaches o en o e look nonlinea
op imiza ion wi hin esidual blocks. This wo k
in oduces SR-ResNe , which in eg a es
Zou endijk’s Me hod o op imize weigh upda es,
ensu ing s able con e gence and imp o ed
classi ica ion accu acy. By add essing a c i ical gap
in con e gence-op imized deep lea ning, SR-
ResNe p o ides a obus solu ion ailo ed o
melanoma de ec ion, ou pe o ming adi ional
models and o e ing a scalable, p ecise diagnos ic
amewo k sui able o clinical deploymen .
2. LITERATURE REVIEW
Ensemble Embeddings App oach [8] has
explo ed melanoma de ec ion by in eg a ing an
ensemble o machine lea ning models wi h deep
ea u e embeddings, demons a ing signi ican
imp o emen s in accu acy. The s udy highligh s he
ad an ages o combining adi ional machine
lea ning algo i hms wi h deep lea ning ea u es,
enhancing o e all p edic i e pe o mance. The
Jou nal o Theo e ical and Applied In o ma ion Technology
31s May 2025. Vol.103. No.10
© Li le Lion Scien i ic
ISSN: 1992-8645 www.ja i .o g E-ISSN: 1817-3195
4193
p oposed ensemble me hod mi iga es o e i ing
and inc eases obus ness, making i a p omising
app oach o melanoma de ec ion in clinical
se ings. MuSClD Sys em [9] has in oduced Mul i-
si e C oss-o gan Calib a ed Deep Lea ning
(MuSClD), an au oma ed sys em o diagnosing
non-melanoma skin cance . The esea ch
emphasizes he impo ance o calib a ing deep
lea ning models ac oss mul iple si es o imp o e
gene alizabili y. The p oposed model le e ages
c oss-o gan calib a ion o enhance he accu acy o
diagnosis, add essing he challenges associa ed
wi h a ying da a dis ibu ions ac oss di e en
clinical se ings. In e p e able DL Sys em [10] ha e
de eloped an in e p e able deep lea ning sys em o
mul i-class segmen a ion and classi ica ion o non-
melanoma skin cance . Thei app oach combines
deep lea ning wi h in e p e abili y echniques,
allowing clinicians o unde s and he decision-
making p ocess o he model. The s udy
demons a es ha in e p e able models can achie e
high accu acy while p o iding insigh s in o he
model's beha io , which is c ucial o clinical
adop ion.
G asshoppe DL Hyb id [11] ha e
p oposed a hyb id G asshoppe op imiza ion
algo i hm combined wi h deep lea ning o skin
lesion segmen a ion and melanoma classi ica ion.
The esea ch in eg a es e olu iona y algo i hms
wi h deep lea ning o op imize he segmen a ion
p ocess, imp o ing classi ica ion accu acy. The
hyb id app oach add esses he limi a ions o
adi ional deep lea ning models in segmen ing
complex skin lesions, p o iding a mo e eliable ool
o melanoma diagnosis. Hype spec al Signa u e
Lea ning [12]has been explo ed o classi y ac inic
ke a osis and non-melanoma skin cance s using
nea -in a ed hype spec al signa u es. This
esea ch shows ha hype spec al imaging in
conjunc ion wi h machine lea ning can e ec i ely
di e en ia e be ween a ious skin lesions. Thei
echnology p o ides an al e na i e o in asi e
p ocedu es o ea ly iden i ica ion o skin cance ,
which has he po en ial o g ea ly imp o e
diagnos ic accu acy.Deep Lea ning Classi ie
[13]s udied he e icacy o se e al deep lea ning
a chi ec u es on da ase s consis ing o skin lesion
in o ma ion in o de o iden i y and ca ego ize skin
cance s. Findings show ha deep lea ning
algo i hms can de ec skin cance mo e accu a ely
han con en ional app oaches, sugges ing ha hey
may e en ually eplace hem. I del es in o he
di icul ies encoun e ed by deep lea ning models,
including o e i ing and he equi emen o
ex ensi e da ase s.
Op imized DL Segmen a ion
[14]highligh s he impo ance o accu a e
segmen a ion o be e diagnosis accu acy by
op imizing deep lea ning me hods o skin lesion
segmen a ion and skin cance de ec ion. Thei
esea ch inco po a es ad anced deep-lea ning
a chi ec u es ailo ed o handle he unique
challenges posed by skin lesion images. The s udy
demons a es ha op imized deep-lea ning models
can achie e supe io pe o mance in skin cance
de ec ion, pa icula ly in challenging cases. Hyb id
DL F amewo k [15] ha e in oduced a hyb id deep
lea ning amewo k o p edic ing skin cance ,
combining a ious deep lea ning echniques o
imp o e accu acy. The amewo k in eg a es
con olu ional neu al ne wo ks wi h o he deep
lea ning me hods, esul ing in a obus model
capable o handling di e se skin lesion images.
Thei app oach o e s a comp ehensi e solu ion o
skin cance p edic ion, enhancing he model's
gene alizabili y ac oss di e en da ase s.
Pola ime ic ML Classi ie [16] has used
pola ime ic imaging wi h machine lea ning o
classi y non-melanoma skin cance in mice issues.
The s udy le e ages op ical pa ame e s de i ed
om pola ime ic imaging o imp o e classi ica ion
accu acy. Thei indings sugges combining
pola ime ic imaging wi h machine lea ning
p o ides a no el app oach o skin cance diagnosis,
po en ially o e ing mo e accu a e esul s han
adi ional imaging me hods. T ans e Lea ning
Hyb id[17]iden i y melanoma skin cance by using
ans e lea ning o segmen a ion in conjunc ion
wi h hyb id classi ica ion. This s udy aims o
imp o e melanoma de ec ion sys ems by u ilizing
p e- ained models, especially in cases when
labeled da a is ew. The wo k shows ha melanoma
diagnosis may be much imp o ed using ans e
lea ning and hyb id classi ica ion algo i hms, which
makes i a iable ool o clinical si ua ions.
Ensemble T ans e Lea ning (ETL) [18]is
u ilizing deep ans e lea ning and ensemble
machine lea ning in a combined e o o classi y
melanoma. In o de o imp o e classi ica ion
accu acy, he s udy in es iga es using deep lea ning
in conjunc ion wi h se e al machine lea ning
models. The ensemble app oach add esses he
challenges o o e i ing and model obus ness,
p o iding a mo e eliable me hod o melanoma
de ec ion. Thei indings indica e combining
adi ional and deep lea ning me hods can lead o
Jou nal o Theo e ical and Applied In o ma ion Technology
31s May 2025. Vol.103. No.10
© Li le Lion Scien i ic
ISSN: 1992-8645 www.ja i .o g E-ISSN: 1817-3195
4194
supe io classi ica ion pe o mance.DL Me a-
analysis (DLMA)[19]ha e e iewed and analyzed
deep lea ning algo i hms o de moscopy-based
melanoma diagnosis in a sys ema ic way. In o de
o assess deep lea ning's e icacy in melanoma
diagnosis, hei s udy combines esul s om se e al
in es iga ions. While deep lea ning models ha e
demons a ed p omising esul s in inc easing
diagnosis accu acy, he s udy no es ha he e is s ill
a need o mo e esea ch in o issues like da a
quali y and model in e p e abili y.Bio-inspi ed
Op imiza ion plays a signi ican ole in di e en
esea ch o achie e he bes esul [20]-[53]. E en
in skin cance de ec ion also, bio-inspi ed
op imiza ion can be applied.
3. SELF-RELIANT RESIDUAL NETWORK
Sel -Relian ResNe (SR-ResNe )
ep esen s a subs an ial ad ancemen o e
adi ional ResNe a chi ec u es by
in eg a ingZou endijk's Me hod, a sophis ica ed
nonlinea op imiza ion echnique. By le e aging
Zou endijk's Me hod, SR-ResNe op imizes he
weigh upda e p ocess du ing aining, signi ican ly
imp o ing con e gence s abili y and compu a ional
e iciency. This in eg a ion e ec i ely mi iga es
issues such as g adien anishing and o e i ing,
which a e common challenges in deep lea ning
models, pa icula ly in complex asks like
melanoma de ec ion. The enhanced op imiza ion
p ocess enables SR-ResNe o achie e mo e p ecise
and eliable p edic ions o melanoma. I is a
powe ul ool in applica ions ha demand high
accu acy and obus ness in de ec ing his agg essi e
o m o skin cance .
3.1. Inpu Laye
The i s s ep in SR-ResNe in ol es he
inpu laye , whe e he aw da a, ypically an image,
is ed in o he ne wo k. The inpu image can be
ep esen ed as a enso o dimensions
whe e ep esen s he heigh ,
ep esen s he wid h and deno es he numbe o
channels, usually co esponding o he colo
channels in an RGB image . The
ma hema ical o mula ion o he inpu enso can be
desc ibed as:
o
(1)
In he ini ial con olu ional laye , he
inpu enso unde goes a con olu ion ope a ion
wi h a se o il e s o ke nels, deno ed as ,
whe e is a ou -dimensional enso o
dimensions He e, and
deno e he heigh and wid h o he il e ,
espec i ely, and ep esen s he numbe o il e s
applied. The con olu ion ope a ion be ween and
can be ma hema ically ep esen ed as:
(2)
whe e is he ou pu ea u e map esul ing om
he con olu ion ope a ion a posi ion . The
abo e equa ion compu es he weigh ed sum o he
inpu pixel alues and he co esponding il e
weigh s, p oducing an ou pu ea u e map o
educed dimensions, which cap u es he essen ial
ea u es om he inpu image.
A nonlinea i y is in oduced in o he
model a e he con olu ion p ocess by applying an
ac i a ion unc ion elemen -wise o he ou pu
ea u e map . The ma hema ical exp ession o he
mos used ac i a ion unc ion, he Rec i ied Linea
Uni (ReLU):
(3)
whe e is he ac i a ed ou pu a posi ion .
This ope a ion e ains posi i e alues and se s all
nega i e alues o ze o, which helps o p e en he
anishing g adien p oblem.
In speci ic a chi ec u es, a max-pooling
ope a ion ollows he ac i a ion unc ion, educing
he spa ial dimensions o he ea u e maps by
selec ing he maximum alue om non-o e lapping
sub egions wi hin he ea u e map. This ope a ion
can be de ined as:
o
(4)
whe e is he pooled ou pu , ep esen s
he size o he pooling window, and is he
ac i a ed ea u e map. This pooling p ocess educes
he compu a ional complexi y and p o ides spa ial
in a iance, he eby se ing up he ini ial ea u e
map o u he p ocessing in he subsequen
laye s o SR-ResNe .
3.2. Ini ial Con olu ion and Pooling
The ini ial ea u e map unde goes u he
p ocessing h ough con olu ional laye s wi hin he
esidual block. The ini ial ea u e map ob ained
a e he i s con olu ional and pooling ope a ions,
is passed h ough a se ies o con olu ional laye s
designed o ex ac deepe ea u es. Le ha e
dimensions. The i s con olu ional
Jou nal o Theo e ical and Applied In o ma ion Technology
31s May 2025. Vol.103. No.10
© Li le Lion Scien i ic
ISSN: 1992-8645 www.ja i .o g E-ISSN: 1817-3195
4195
laye wi hin his block applies a se o il e s.
wi h dimensions whe e
ep esen s he numbe o il e s. The ope a ion is
de ined as:
(5
)
whe e ep esen s he ou pu ea u e map
a e he i s con olu ional ope a ion in laye .
A ba ch no maliza ion s ep is applied o
he ou pu ea u e map. , which no malizes he
ou pu o ha e a mean o ze o and a a iance o
one. Le and ep esen he mean and
a iance, espec i ely:
(6)
whe e is a small cons an added o nume ical
s abili y.
Following ba ch no maliza ion, he
ac i a ion unc ion , ypically ReLU, is applied:
(7
)
This p ocess is epea ed in subsequen
con olu ional laye s wi hin he same esidual block.
Conside he nex con olu ional laye wi h
il e s. o dimensions , whe e
is he numbe o il e s in he second
con olu ional laye . The ou pu o his laye is:
(8
)
Ba ch no maliza ion is again applied o
(9)
Following no maliza ion, he ReLU ac i a ion
unc ion is applied:
(10
)
In a s anda d esidual block, hese
ope a ions a e ollowed by adding he o iginal inpu
o he inal ou pu o he con olu ional laye s. In
SR-ResNe , his s ep inco po a es Zou endijk's
Me hod o op imiza ion.The esidual unc ion is
deno ed as:
(11)
The inpu is added o he esidual
unc ion. yielding he ou pu o he
esidual block:
(12)
Zou endijk's Me hod is hen applied o
op imize , whe e he easible di ec ion
and s ep size a e calcula ed. The upda e
s ep o he pa ame e s in ol es mo ing in he
di ec ion by a s ep size
(13)
The op imized ou pu becomes
he inpu o he nex block in he SR-ResNe ,
ensu ing e icien aining and enhanced
con e gence h oughou he ne wo k.
3.3. Residual Block wi h Zou endijk's
Op imiza ion
The p ocess con inues wi h s acking
mul iple esidual blocks, each inco po a ing he
enhanced op imiza ion echnique inspi ed by
Zou endijk's Me hod. The inpu o each subsequen
esidual block is he ou pu om he p e ious
block, op imized o imp o e aining s abili y and
con e gence. Le ep esen he ou pu o he
esidual block, which se es as he inpu o he
nex block.The inpu is i s passed h ough a
con olu ional laye wi h il e s de ined by he
dimensions whe e is he
numbe o channels in and is he numbe o
il e s. The ope a ion is exp essed as:
(14
)
Following his con olu ion, he ba ch
no maliza ion s ep is applied o he ou pu ea u e
map. wi h he mean and a iance
calcula ed as:
(15)
The ou pu is hen passed h ough he
ac i a ion unc ion, ypically ReLU:
(16
)

Jou nal o Theo e ical and Applied In o ma ion Technology
31s May 2025. Vol.103. No.10
© Li le Lion Scien i ic
ISSN: 1992-8645 www.ja i .o g E-ISSN: 1817-3195
4196
The p ocess con inues wi h ano he
con olu ional laye , his ime using il e s. wi h
dimensions whe e
ep esen s he numbe o il e s in he nex laye .
The con olu ion ope a ion is gi en by:
(1
7)
Ba ch no maliza ion is again applied:
(18)
The ac i a ed ou pu is hen:
(19
)
This ou pu se es as he esidual
unc ion o he nex laye . To main ain
con inui y, he o iginal inpu is added o his
esidual unc ion, o ming he inpu o he nex
block:
(20)
A his s age, Zou endijk's Me hod is
applied again. The objec i e unc ion
ep esen ing he loss, is op imized by de e mining a
easible di ec ion and an op imal s ep size
(21)
The op imized se o pa ame e s
upda es he weigh s wi hin he block, ensu ing ha
he ne wo k con inues o lea n e icien ly. The
ou pu becomes he inpu o he subsequen
esidual block, p ese ing he g adien low and
enhancing he o e all ne wo k pe o mance.By
epea ing his p ocess, SR-ResNe cons uc s a deep
ne wo k whe e each esidual block is op imized o
imp o ed con e gence, leading o be e model
accu acy and e iciency in lea ning complex
pa e ns.
3.4. S acking SR-ResNe Blocks
The ou pu om he se ies o esidual
blocks is p ocessed h ough global a e age pooling,
ully connec ed laye s, and he inal op imiza ion
s eps o p oduce he ne wo k's ou pu . The inpu o
his s ep is he ou pu om he las esidual block,
deno ed as whe e is he index o he inal
esidual block in he ne wo k.The global a e age
pooling ope a ion is applied o which has
dimensions The ea u e maps a e
agg ega ed in o a single alue using he global
a e age pooling me hod, which a e ages all he
spa ial componen s.This ope a ion is
ma hema ically exp essed as:
(22)
whe e is he pooled ou pu co esponding o
he ea u e map, and The esul
is a ec o o leng h.
This pooled ec o is hen passed o a
ully connec ed laye wi h a weigh ma ix and
bias ec o The ully connec ed laye compu es
he ou pu as:
(23)
whe e is he ou pu o he neu on in he ully
connec ed laye , and wi h being
he numbe o neu ons in he ully connec ed laye .
A e he ully connec ed laye , he ou pu s
a e ans o med in o p obabili ies using a so max
ac i a ion unc ion. The so max unc ion o he
h class is de ined as:
(24)
whe e ep esen s he p edic ed p obabili y o he
class.
The loss unc ion is de e mined
du ing aining by using c oss-en opy loss, which
quan i ies he disc epancy be ween he ac ual labels
and he an icipa ed p obabili ies .The c oss-
en opy loss is gi en by:
(25)
whe e is he g ound u h label o he class.
To op imize he ne wo k, Zou endijk's
Me hod is applied, which in ol es de e mining he
easible di ec ion and he op imal s ep size o
minimizing he loss unc ion. The weigh upda es
o he ully connec ed laye a e pe o med as:
(26)
Jou nal o Theo e ical and Applied In o ma ion Technology
31s May 2025. Vol.103. No.10
© Li le Lion Scien i ic
ISSN: 1992-8645 www.ja i .o g E-ISSN: 1817-3195
4197
whe e ep esen s he weigh s a i e a ion and
is he easible di ec ion de e mined by
Zou endijk's Me hod.
The back p opaga ion p ocess upda es he
pa ame e s h oughou he ne wo k, including he
weigh s in he con olu ional laye s and ully
connec ed laye s, ensu ing he op imiza ion o he
o e all objec i e unc ion The inal ou pu o
SR-ResNe is he class wi h he highes p obabili y
om he so max laye , ep esen ing he ne wo k's
p edic ion.
3.5. Bo leneck Blocks
The p ima y ocus is op imizing he en i e
ne wo k using Zou endijk's Me hod, which is
in eg a ed wi h he backp opaga ion algo i hm. The
op imiza ion p ocess begins wi h calcula ing
g adien s o all ne wo k pa ame e s, ollowed by
de e mining easible di ec ions and op imal s ep
sizes o weigh upda es.To ini ia e he
op imiza ion, he loss unc ion whe e
ep esen s he en i e se o ne wo k pa ame e s, is
compu ed. Each pa ame e is used o ep esen
he g adien o he loss unc ion.
(27)
The chain ule is used o compu e he
g adien . o e e y laye in he ne wo k.
To minimize he loss, he pa ame e s migh be
modi ied in he di ec ion indica ed by his
g adien .The g adien a laye o weigh is
gi en by:
(28)
The ou pu o he laye be o e adding an
ac i a ion unc ion is ep esen ed by .Once he
g adien s a e calcula ed, Zou endijk's Me hod is
applied o de e mine a easible di ec ion. o
each pa ame e upda e. This in ol es sol ing he
op imiza ion p oblem:
(29)
(30)
whe e ep esen s any cons ain s on he
pa ame e s and is he s ep size. The di ec ion
is selec ed o ensu e ha he loss dec eases
while main aining easibili y conce ning he
cons ain s. The op imal s ep size is de e mined
by minimizing he loss along he di ec ion :
(31)
The pa ame e s o laye , a e hen
upda ed using he easible di ec ion and he
calcula ed s ep size as ollows:
(32)
This upda e ule is applied o all laye s,
ensu ing ha each se o pa ame e s is op imized
acco ding o Zou endijk's Me hod. The p ocess is
i e a i e, wi h g adien s ecalcula ed a e each
upda e un il con e gence is achie ed, i.e., when he
loss unc ion s abilizes a a minimum o nea -
minimum alue.The back p opaga ion p ocess, in
conjunc ion wi h Zou endijk's Me hod, ensu es ha
he ne wo k's pa ame e s a e upda ed in he
di ec ion ha educes he loss and op imally wi hin
he easible egion de ined by any cons ain s. This
esul s in a obus and e icien lea ning p ocess o
he en i e SR-ResNe a chi ec u e.
3.6. Downsampling
The ocus is on he i e a i e p ocess o
e ining he model h ough epea ed applica ion o
Zou endijk's Me hod ac oss mul iple aining
epochs. The p ocess ensu es ha he model
p og essi ely mo es close o he global o local
minimum o he loss unc ion while
espec ing any cons ain s imposed on he
pa ame e s. Gi en he pa ame e se a epoch
he objec i e is o u he minimize he loss
unc ion by upda ing he pa ame e s h ough a
se ies o i e a ions. To ge he loss unc ion's
g adien conce ning he pa ame e s, one uses he
ollowing o mula:
(33)
Zou endijk's Me hod is hen applied o
de e mine he easible di ec ion. ha minimizes
he objec i e unc ion while ensu ing he
pa ame e s emain wi hin he easible egion. This
op imiza ion p oblem can be exp essed as:
(34)
(35)
Jou nal o Theo e ical and Applied In o ma ion Technology
31s May 2025. Vol.103. No.10
© Li le Lion Scien i ic
ISSN: 1992-8645 www.ja i .o g E-ISSN: 1817-3195
4198
whe e ep esen s any cons ain s, such as non-
nega i i y o boundedness, ha mus be sa is ied by
he pa ame e s .
To e ine he pa ame e s, he op imal s ep
size is de e mined by minimizing he loss along
he di ec ion :
(36)
The pa ame e s a e hen upda ed o he
nex i e a ion using he calcula ed di ec ion
and s ep size:
(37)
As aining p og esses h ough mul iple
epochs, he g adien is ecalcula ed
a e each pa ame e upda e, ensu ing ha he
pa ame e s con inually mo e in he di ec ion mos
e ec i ely educes he loss unc ion.The i e a ion
p ocess con inues wi h upda ed
g adien s. and di ec ions o each
subsequen epoch:
(38)
(39)
The e inemen p ocess also in ol es
ensu ing ha he cons ain s emain
sa is ied. This equi es ecalcula ing he easible
egion a each i e a ion and adjus ing he s ep
size. acco dingly o main ain easibili y:
(40)
This i e a i e p ocess con inues ac oss
mul iple epochs un il he loss unc ion eaches
a s able minimum, indica ing ha he model
pa ame e s ha e con e ged o op imal alues. The
epea ed applica ion o Zou endijk's Me hod
h oughou his p ocess ensu es ha he pa ame e
upda es a e e ec i e in educing he loss and obus
agains po en ial cons ain iola ions, leading o a
well-op imized model in he SR-ResNe
amewo k.
3.7. Global A e age Pooling
The ocus is on e alua ing and adjus ing
he model du ing he aining p ocess o ensu e he
con e gence and s abili y o he op imiza ion. This
s ep in ol es moni o ing he loss unc ion,
changing he lea ning a e, and applying
egula iza ion echniques o enhance he model's
gene aliza ion capabili ies.The p ima y objec i e in
his s ep is o minimize he loss unc ion by
e ining he pa ame e s while ensu ing ha he
model does no o e i he aining da a. The
p ocess begins by e alua ing he loss unc ion a e
each aining epoch. The loss unc ion a epoch is
deno ed as:
(41
)
whe e ep esen s he numbe o aining samples,
is he ue label, and is p edic ed
p obabili y o he sample.
The loss unc ion's g adien conce ning he
pa ame e s may be de e mined using he
ollowing o mula:
(42
)
Applying egula iza ion me hods, such as
L2 egula iza ion, helps a oid o e i ing and
enhances gene aliza ion.The L2 egula iza ion e m
is added o he loss unc ion:
(43)
whe e con ols he penal y's in ensi y as a
egula iza ion pa ame e .
The o al loss wi h egula iza ion is
minimized by upda ing he pa ame e s using he
g adien descen me hod wi h a lea ning a e.
ha is po en ially adjus ed du ing aining:
(44)
To ensu e s able con e gence, he lea ning
a e may be adjus ed based on he e alua ion o
he loss unc ion o e successi e epochs. I he loss
does no dec ease as expec ed, he lea ning a e is
educed:
wi h (45)
This s ep also in ol es moni o ing he
model's pe o mance on a alida ion se . The
alida ion loss is calcula ed simila ly o he
aining loss. I he alida ion loss s a s inc easing
while he aining loss dec eases, ea ly s opping
c i e ia may be applied o p e en o e i ing:
Jou nal o Theo e ical and Applied In o ma ion Technology
31s May 2025. Vol.103. No.10
© Li le Lion Scien i ic
ISSN: 1992-8645 www.ja i .o g E-ISSN: 1817-3195
4199
I , hen s op aining (46)
Con inuous e alua ion o he loss unc ion,
adjus men o he lea ning a e, and applica ion o
egula iza ion ensu e ha SR-ResNe con e ges o
a well-gene alized solu ion ha e ec i ely
pe o ms on aining and unseen da a.
3.8. Fully Connec ed Laye
In Fully Connec ed Laye , he p ocess
in ol es inal adjus men s and he e alua ion o he
ained model's pe o mance on he es da a. This
s ep ensu es he model gene alizes well o new,
unseen da a and e alua es i s obus ness ac oss
di e en scena ios.The ained model pa ame e s
ob ained a e he aining p ocess, a e used o
make p edic ions on he es da ase . Fo each es
sample he model compu es he p edic ed
p obabili y o he a ge class using he
so max unc ion:
(47)
whe e deno es he numbe o classes and
ep esen s he ou pu o he inal laye o he
es sample.
The p edic ed class label o each es
sample is de e mined by selec ing he class wi h he
highes p edic ed p obabili y:
(48)
Nex , measu es like F1-sco e, ecall,
accu acy, and p ecision assess he model's o e all
pe o mance. Using he numbe o es samples
di ided by he numbe o p ope ly p edic ed labels,
we can de e mine he accu acy:
(49)
The indica o unc ion is de ined as 1
when he p edic ed label is iden ical o he eal label
and 0 o he wise, wi h being he o al numbe
o es samples.
P ecision, ecall, and F1-sco e a e calcula ed o
p o ide a mo e de ailed analysis o he model's
pe o mance, pa icula ly in imbalanced da ase s.
P ecision o a class is de ined as:
(50)
whe e and deno e he ue posi i es and
alse posi i es o class , espec i ely. Recall o
class is de ined as:
(51)
whe e deno es he alse nega i es o class .
The F1-sco e o class is he ha monic mean o
p ecision and ecall:
(52)
To assess he o e all pe o mance ac oss
all classes, he mac o-a e aged F1-sco e is
calcula ed:
(53)
A e e alua ing he pe o mance on he
es se , i he model's pe o mance me ics mee he
desi ed c i e ia, he model is conside ed eady o
deploymen . Howe e , i he pe o mance is
subop imal, he aining p ocess may be e isi ed,
and adjus men s o he lea ning a e, egula iza ion,
o a chi ec u e may be made.
3.9. So max Ac i a ion
SR-ResNe unde goes e inemen h ough
ine- uning, which aims o op imize he ne wo k's
pe o mance on a speci ic ask o da ase . Fine-
uning in ol es e aining some o all ne wo k
laye s using a lowe lea ning a e while po en ially
adjus ing o he hype pa ame e s o enhance
pe o mance. The ine- uning p ocess begins by e-
e alua ing he loss unc ion on he a ge
da ase , whe e ep esen s he cu en se o
ne wo k pa ame e s. A new calcula ion is made o
he loss unc ion's g adien conce ning hese
pa ame e s, and i is:
(54)
In ine- uning, he lea ning a e is
ypically educed compa ed o he ini ial aining
phase o allow o mo e p ecise weigh adjus men s.
The lea ning a e a i e a ion du ing ine- uning is
exp essed as:
(55)
whe e is he ini ial lea ning a e used du ing ine-
uning, and is a decay ac o such ha
Jou nal o Theo e ical and Applied In o ma ion Technology
31s May 2025. Vol.103. No.10
© Li le Lion Scien i ic
ISSN: 1992-8645 www.ja i .o g E-ISSN: 1817-3195
4206
AI, and op imizing deploymen o clinical
in eg a ion.
REFERENCES
[1] K. Se hanan e al., “Double AMIS-ensemble
deep lea ning o skin cance classi ica ion,”
Expe Sys . Appl., ol. 234, p. 121047, 2023,
doi:
h ps://doi.o g/10.1016/j.eswa.2023.121047.
[2] W. Abbes and D. Sellami, “Deep Neu al
Ne wo ks o Melanoma De ec ion om
Op ical S anda d Images using T ans e
Lea ning,” P ocedia Compu . Sci., ol. 192,
pp. 1304–1312, 2021, doi:
h ps://doi.o g/10.1016/j.p ocs.2021.08.134.
[3] F. Alenezi, A. A mghan, and K. Pola , “A
mul i-s age melanoma ecogni ion amewo k
wi h deep esidual neu al ne wo k and
hype pa ame e op imiza ion-based decision
suppo in de moscopy images,” Expe Sys .
Appl., ol. 215, p. 119352, 2023, doi:
h ps://doi.o g/10.1016/j.eswa.2022.119352.
[4] S. Kad y, E. Ve dú, R. Damase icius, L.
Abualigah, V. Singh, and V. Rajinikan h,
“CNN segmen a ion o skin melanoma in p e-
p ocessed de moscopy images,” P ocedia
Compu . Sci., ol. 235, pp. 2775–2782, 2024,
doi:
h ps://doi.o g/10.1016/j.p ocs.2024.04.262.
[5] Q. Liu, H. Kawashima, and A. Rezaei so la,
“An op imal me hod o melanoma de ec ion
om de moscopy images using ein o cemen
lea ning and suppo ec o machine op imized
by enhanced ish mig a ion op imiza ion
algo i hm,” Heliyon, ol. 9, no. 10, p. e21118,
2023, doi:
h ps://doi.o g/10.1016/j.heliyon.2023.e21118.
[6] L. Ande sson e al., “Poly(e hylene glycol)-
poly(es e -ca bona e) block copolyme s
ca ying PEG-pep idyl-doxo ubicin pendan
side chains: Syn hesis and e alua ion as
an icance conjuga es,” Biomac omolecules,
ol. 6, no. 2, pp. 914–926, 2005, doi:
10.1021/bm049381p.
[7] Ç. Suiçmez, H. Tolga Kah aman, A. Suiçmez,
C. Yılmaz, and F. Balcı, “De ec ion o
melanoma wi h hyb id lea ning me hod by
emo ing hai om de moscopic images using
image p ocessing echniques and wa ele
ans o m,” Biomed. Signal P ocess. Con ol,
ol. 84, p. 104729, 2023, doi:
10.1016/j.bspc.2023.104729.
[8] S. Ghosh, S. Dha , R. Yoddha, S. Kuma , A.
K. Thaku , and N. D. Jana, “Melanoma Skin
Cance De ec ion Using Ensemble o Machine
Lea ning Models Conside ing Deep Fea u e
Embeddings,” P ocedia Compu . Sci., ol. 235,
pp. 3007–3015, 2024, doi:
h ps://doi.o g/10.1016/j.p ocs.2024.04.284.
[9] Y. Zhou e al., “Mul i-si e c oss-o gan
calib a ed deep lea ning (MuSClD):
Au oma ed diagnosis o non-melanoma skin
cance ,” Med. Image Anal., ol. 84, p. 102702,
2023, doi:
h ps://doi.o g/10.1016/j.media.2022.102702.
[10] S. M. Thomas, J. G. Le e e, G. Bax e , and N.
A. Hamil on, “In e p e able deep lea ning
sys ems o mul i-class segmen a ion and
classi ica ion o non-melanoma skin cance ,”
Med. Image Anal., ol. 68, p. 101915, 2021,
doi:
h ps://doi.o g/10.1016/j.media.2020.101915.
[11] P. Thapa , M. Rakh a, M. Alsaadi, A. Qu aishi,
A. Deka, and J. V. Naga Ramesh, “A hyb id
G asshoppe op imiza ion algo i hm o skin
lesion segmen a ion and melanoma
classi ica ion using deep lea ning,” Heal hc.
Anal., ol. 5, p. 100326, 2024, doi:
h ps://doi.o g/10.1016/j.heal h.2024.100326.
[12] L. A. Cou enay, I. Ba be o-Ga cía, S.
Ma ínez-Las as, S. Del Pozo, M. Co al, and
D. González-Aguile a, “Using compu a ional
lea ning o non-melanoma skin cance and
ac inic ke a osis nea -in a ed hype spec al
signa u e classi ica ion,” Pho odiagnosis
Pho odyn. The ., ol. 49, p. 104269, 2024, doi:
h ps://doi.o g/10.1016/j.pdpd .2024.104269.
[13] C. Ka i ha, S. P iyanka, M. P. Kuma , and V.
Kusuma, “Skin Cance De ec ion and
Classi ica ion using Deep Lea ning
Techniques,” P ocedia Compu . Sci., ol. 235,
pp. 2793–2802, 2024, doi:
h ps://doi.o g/10.1016/j.p ocs.2024.04.264.
[14] R. R. Babu and F. M. Philip, “Op imized deep
lea ning o skin lesion segmen a ion and skin
cance de ec ion,” Biomed. Signal P ocess.
Con ol, ol. 95, p. 106292, 2024, doi:
h ps://doi.o g/10.1016/j.bspc.2024.106292.
[15] E. Fa ea, R. A. A. Saleh, H. AbuAlkebash, A.
A. R. Fa ea, and M. A. Al-an a i, “A hyb id
deep lea ning skin cance p edic ion
amewo k,” Eng. Sci. Technol. an In . J., ol.
57, p. 101818, 2024, doi:
h ps://doi.o g/10.1016/j.jes ch.2024.101818.
[16] T.-T.-H. Pham, T.-N. Luu, T.-V. Nguyen, N.-
T. Huynh, Q.-H. Phan, and T.-H. Le,
“Pola ime ic imaging combining op ical
pa ame e s o classi ica ion o mice non-
melanoma skin cance issue using machine

Jou nal o Theo e ical and Applied In o ma ion Technology
31s May 2025. Vol.103. No.10
© Li le Lion Scien i ic
ISSN: 1992-8645 www.ja i .o g E-ISSN: 1817-3195
4207
lea ning,” Heliyon, ol. 9, no. 11, p. e22081,
2023, doi:
h ps://doi.o g/10.1016/j.heliyon.2023.e22081.
[17] R. Dandu, M. Vinayaka Mu hy, and Y. B.
Ra i Kuma , “T ans e lea ning o
segmen a ion wi h hyb id classi ica ion o
De ec Melanoma Skin Cance ,” Heliyon, ol.
9, no. 4, p. e15416, 2023, doi:
h ps://doi.o g/10.1016/j.heliyon.2023.e15416.
[18] M. Roshni Thanka e al., “A hyb id app oach
o melanoma classi ica ion using ensemble
machine lea ning echniques wi h deep ans e
lea ning,” Compu . Me hods P og ams
Biomed. Upda ., ol. 3, p. 100103, 2023, doi:
h ps://doi.o g/10.1016/j.cmpbup.2023.100103.
[19] Z. Ye e al., “Deep lea ning algo i hms o
melanoma de ec ion using de moscopic
images: A sys ema ic e iew and me a-
analysis,” A i . In ell. Med., ol. 155, p.
102934, 2024, doi:
h ps://doi.o g/10.1016/j.a med.2024.102934.
[20] B. Suchi a, J. Ramkuma , and R. Ka hikeyan,
“F og Leap Inspi ed Op imiza ion-Based
Ex eme Lea ning Machine Fo Accu a e
Classi ica ion O La en Au oimmune Diabe es
In Adul s (LADA),” J. Theo . Appl. In .
Technol., ol. 103, no. 2, pp. 472–494, 2025,
[Online]. A ailable:
h ps://www.scopus.com/inwa d/ eco d.u i?eid
=2-s2.0-
85217140979&pa ne ID=40&md5=9540433c
16d5 0 6c2de4b8c43a4812
[21] J. Ramkuma , V. Vala ma hi, and R.
Ka hikeyan, “Op imizing Quali y o Se ice
and Ene gy E iciency in Haza dous D one
Ad-Hoc Ne wo ks (DANET) Using King ishe
Rou ing P o ocol (KRP),” In . J. Eng. T ends
Technol., ol. 73, no. 1, pp. 410–430, 2025,
doi: 10.14445/22315381/IJETT-V73I1P135.
[22] R. Jagana han, S. Meh a, and R. K ishan,
“P e ace,” Bio-Inspi ed In ell. Sma Decis.,
pp. xix–xx, 2024, [Online]. A ailable:
h ps://www.scopus.com/inwa d/ eco d.u i?eid
=2-s2.0-
85195725049&pa ne ID=40&md5=7a2aa7adc
005662eebc12e 82e3bd19
[23] J. Ramkuma , B. Va un, V. Vala ma hi, D. R.
Medhunhashini, and R. Ka hikeyan, “Jagua -
Based Rou ing P o ocol (J p) Fo Imp o ed
Reliabili y And Reduced Packe Loss In D one
Ad-Hoc Ne wo ks (DANET),” J. Theo . Appl.
In . Technol., ol. 103, no. 2, pp. 696–713,
2025, [Online]. A ailable:
h ps://www.scopus.com/inwa d/ eco d.u i?eid
=2-s2.0-
85217213044&pa ne ID=40&md5=e38a375e
46c 43c95d6702a3585a7073
[24] J. Ramkuma and R. Vadi el, “Mul i-Adap i e
Rou ing P o ocol o In e ne o Things based
Ad-hoc Ne wo ks,” Wi el. Pe s. Commun., ol.
120, no. 2, pp. 887–909, Ap . 2021, doi:
10.1007/s11277-021-08495-z.
[25] A. Sen hilkuma , J. Ramkuma , M. Linga aj,
D. Jaya aj, and B. Su eshkuma , “Minimizing
Ene gy Consump ion in Vehicula Senso
Ne wo ks Using Relen less Pa icle Swa m
Op imiza ion Rou ing,” In . J. Compu .
Ne wo ks Appl., ol. 10, no. 2, pp. 217–230,
2023, doi: 10.22247/ijcna/2023/220737.
[26] S. P. P iyadha shini and J. Ramkuma ,
“Mappings O Pli hogenic Cubic Se s,”
Neu osophic Se s Sys ., ol. 79, pp. 669–685,
2025, doi: 10.5281/zenodo.14607210.
[27] J. Ramkuma and R. Vadi el, “Imp o ed Wol
p ey inspi ed p o ocol o ou ing in cogni i e
adio Ad Hoc ne wo ks,” In . J. Compu .
Ne wo ks Appl., ol. 7, no. 5, pp. 126–136,
2020, doi: 10.22247/ijcna/2020/202977.
[28] R. Jagana han and R. Vadi el, “In elligen Fish
Swa m Inspi ed P o ocol (IFSIP) o Dynamic
Ideal Rou ing in Cogni i e Radio Ad-Hoc
Ne wo ks,” In . J. Compu . Digi . Sys ., ol. 10,
no. 1, pp. 1063–1074, 2021, doi:
10.12785/ijcds/100196.
[29] J. Ramkuma and R. Vadi el, “CSIP—cuckoo
sea ch inspi ed p o ocol o ou ing in
cogni i e adio ad hoc ne wo ks,” in Ad ances
in In elligen Sys ems and Compu ing, Sp inge
Ve lag, 2017, pp. 145–153. doi: 10.1007/978-
981-10-3874-7_14.
[30] S. P. Gee ha, N. M. S. Sunda i, J. Ramkuma ,
and R. Ka hikeyan, “Ene gy E icien Rou ing
In Quan um Flying Ad Hoc Ne wo k (Q-Fane )
Using Mamdani Fuzzy In e ence Enhanced
Dijks a’s Algo i hm (MFI-EDA),” J. Theo .
Appl. In . Technol., ol. 102, no. 9, pp. 3708–
3724, 2024, [Online]. A ailable:
h ps://www.scopus.com/inwa d/ eco d.u i?eid
=2-s2.0-
85197297302&pa ne ID=40&md5=72d51668
bee6239 09a59d2694d 67d6
[31] M. P. Swapna, J. Ramkuma , and R.
Ka hikeyan, “Ene gy-Awa e Reliable Rou ing
wi h Blockchain Secu i y o He e ogeneous
Wi eless Senso Ne wo ks,” in Lec u e No es
in Ne wo ks and Sys ems, V. Goa , M. Ku i, R.
Kuma , and T. Senjyu, Eds., Sp inge Science
and Business Media Deu schland GmbH, 2025,
pp. 713–723. doi: 10.1007/978-981-97-6106-
7_43.
Jou nal o Theo e ical and Applied In o ma ion Technology
31s May 2025. Vol.103. No.10
© Li le Lion Scien i ic
ISSN: 1992-8645 www.ja i .o g E-ISSN: 1817-3195
4208
[32] J. Ramkuma , K. S. Jeen Ma seline, and D. R.
Medhunhashini, “Relen less Fi e ly
Op imiza ion-Based Rou ing P o ocol
(RFORP) o Secu ing Fin ech Da a in IoT-
Based Ad-Hoc Ne wo ks,” In . J. Compu .
Ne wo ks Appl., ol. 10, no. 4, pp. 668–687,
2023, doi: 10.22247/ijcna/2023/223319.
[33] K. S. J. Ma seline, J. Ramkuma , and D. R.
Medhunhashini, “Sophis ica ed Kalman
Fil e ing-Based Neu al Ne wo k o Analyzing
Sen imen s in Online Cou ses,” in Sma
Inno a ion, Sys ems and Technologies, A. K.
Somani, A. Mund a, R. K. Gup a, S.
Bha acha ya, and A. P. Mazumda , Eds.,
Sp inge Science and Business Media
Deu schland GmbH, 2024, pp. 345–358. doi:
10.1007/978-981-97-3690-4_26.
[34] J. Ramkuma and R. Vadi el, “Whale
op imiza ion ou ing p o ocol o minimizing
ene gy consump ion in cogni i e adio wi eless
senso ne wo k,” In . J. Compu . Ne wo ks
Appl., ol. 8, no. 4, pp. 455–464, 2021, doi:
10.22247/ijcna/2021/209711.
[35] M. P. Swapna and J. Ramkuma , “Mul iple
Memo y Image Ins ances S a agem o De ec
Fileless Malwa e,” in Communica ions in
Compu e and In o ma ion Science, S.
Rajagopal, K. Popa , D. Me a, and S. Bajeja,
Eds., Sp inge Science and Business Media
Deu schland GmbH, 2024, pp. 131–140. doi:
10.1007/978-3-031-59100-6_11.
[36] D. Jaya aj, J. Ramkuma , M. Linga aj, and B.
Su eshkuma , “AFSORP: Adap i e Fish
Swa m Op imiza ion-Based Rou ing P o ocol
o Mobili y Enabled Wi eless Senso
Ne wo k,” In . J. Compu . Ne wo ks Appl., ol.
10, no. 1, pp. 119–129, 2023, doi:
10.22247/ijcna/2023/218516.
[37] M. Linga aj, T. N. Suguma , C. S. Felix, and J.
Ramkuma , “Que y awa e ou ing p o ocol o
mobili y enabled wi eless senso ne wo k,” In .
J. Compu . Ne wo ks Appl., ol. 8, no. 3, pp.
258–267, 2021, doi:
10.22247/ijcna/2021/209192.
[38] R. Jagana han, S. Meh a, and R. K ishan, Bio-
Inspi ed in elligence o sma decision-
making. IGI Global, 2024. doi:
10.4018/9798369352762.
[39] R. Vadi el and J. Ramkuma , “QoS-enabled
imp o ed cuckoo sea ch-inspi ed p o ocol
(ICSIP) o IoT-based heal hca e applica ions,”
in Inco po a ing he In e ne o Things in
Heal hca e Applica ions and Wea able
De ices, IGI Global, 2019, pp. 109–121. doi:
10.4018/978-1-7998-1090-2.ch006.
[40] J. Ramkuma , R. Ka hikeyan, and M.
Linga aj, “Op imizing IoT-Based Quan um
Wi eless Senso Ne wo ks Using NM-TEEN
Fusion o Ene gy E iciency and Sys ema ic
Go e nance,” in Lec u e No es in Elec ical
Enginee ing, V. Sh i as a a, J. C. Bansal, and
B. K. Panig ahi, Eds., Sp inge Science and
Business Media Deu schland GmbH, 2025, pp.
141–153. doi: 10.1007/978-981-97-6710-6_12.
[41] J. Ramkuma , R. Ka hikeyan, and V.
Vala ma hi, “Alpine Swi Rou ing P o ocol
(ASRP) o S a egic Adap i e Connec i i y
Enhancemen and Boos ed Quali y o Se ice
in D one Ad Hoc Ne wo k (DANET),” In . J.
Compu . Ne wo ks Appl., ol. 11, no. 5, pp.
726–748, 2024, doi: 10.22247/ijcna/2024/45.
[42] P. Menakade i and J. Ramkuma , “Robus
Op imiza ion Based Ex eme Lea ning
Machine o Sen imen Analysis in Big Da a,”
in 2022 In e na ional Con e ence on Ad anced
Compu ing Technologies and Applica ions,
ICACTA 2022, Ins i u e o Elec ical and
Elec onics Enginee s Inc., 2022. doi:
10.1109/ICACTA54488.2022.9753203.
[43] J. Ramkuma , A. Sen hilkuma , M. Linga aj,
R. Ka hikeyan, and L. San hi, “OPTIMAL
APPROACH FOR MINIMIZING DELAYS
IN IOT-BASED QUANTUM WIRELESS
SENSOR NETWORKS USING NM-LEACH
ROUTING PROTOCOL,” J. Theo . Appl. In .
Technol., ol. 102, no. 3, pp. 1099–1111, 2024,
[Online]. A ailable:
h ps://www.scopus.com/inwa d/ eco d.u i?eid
=2-s2.0-
85185481011&pa ne ID=40&md5=b 0 974c
eabc0ad58e589b28797c684
[44] J. Ramkuma and R. Vadi el, “Imp o ed og
leap inspi ed p o ocol (IFLIP) – o ou ing in
cogni i e adio ad hoc ne wo ks (CRAHN),”
Wo ld J. Eng., ol. 15, no. 2, pp. 306–311,
2018, doi: 10.1108/WJE-08-2017-0260.
[45] R. Jagana han, S. Meh a, and R. K ishan,
In elligen Decision Making Th ough Bio-
Inspi ed Op imiza ion. IGI Global, 2024. doi:
10.4018/979-8-3693-2073-0.
[46] R. Jagana han and V. Ramasamy,
“Pe o mance modeling o bio-inspi ed ou ing
p o ocols in Cogni i e Radio Ad Hoc Ne wo k
o educe end- o-end delay,” In . J. In ell. Eng.
Sys ., ol. 12, no. 1, pp. 221–231, 2019, doi:
10.22266/IJIES2019.0228.22.
[47] R. Ka hikeyan and R. Vadi el, “Boos ed
Mu a ed Co ona Vi us Op imiza ion Rou ing
P o ocol (BMCVORP) o Reliable Da a
T ansmission wi h E icien Ene gy
Jou nal o Theo e ical and Applied In o ma ion Technology
31s May 2025. Vol.103. No.10
© Li le Lion Scien i ic
ISSN: 1992-8645 www.ja i .o g E-ISSN: 1817-3195
4209
U iliza ion,” Wi el. Pe s. Commun., ol. 135,
no. 4, pp. 2281–2301, 2024, doi:
10.1007/s11277-024-11155-7.
[48] R. Ka hikeyan and R. Vadi el, “P o icien
Dazzling C ow Op imiza ion Rou ing P o ocol
(PDCORP) o E ec i e Ene gy
Adminis a ion in Wi eless Senso Ne wo ks,”
in IEEE In e na ional Con e ence on
Elec ical, Elec onics, Communica ion and
Compu e s, ELEXCOM 2023, 2023, pp. 1–6.
doi:
10.1109/ELEXCOM58812.2023.10370559.
[49] N. K. Ojha, A. Pandi a, and J. Ramkuma ,
“Cybe secu i y challenges and da k side o AI:
Re iew and cu en s a us,” in Demys i ying he
Da k Side o AI in Business, 2024, pp. 117–
137. doi: 10.4018/979-8-3693-0724-3.ch007.
[50] J. Ramkuma , S. S. Dinaka an, M. Linga aj, S.
Boopalan, and B. Na asimhan, “IoT-Based
Kalman Fil e ing and Pa icle Swa m
Op imiza ion o De ec ing Skin Lesion,” in
Lec u e No es in Elec ical Enginee ing, K.
Mu a i, N. P asad Padhy, and S. Kamalasadan,
Eds., Singapo e: Sp inge Na u e Singapo e,
2023, pp. 17–27. doi: 10.1007/978-981-19-
8353-5_2.
[51] J. Ramkuma , C. Kumu hini, B. Na asimhan,
and S. Boopalan, “Ene gy Consump ion
Minimiza ion in Cogni i e Radio Mobile Ad-
Hoc Ne wo ks using En iched Ad-hoc On-
demand Dis ance Vec o P o ocol,” in 2022
In e na ional Con e ence on Ad anced
Compu ing Technologies and Applica ions,
ICACTA 2022, 2022. doi:
10.1109/ICACTA54488.2022.9752899.
[52] J. Ramkuma , R. Vadi el, and B. Na asimhan,
“Cons ained Cuckoo Sea ch Op imiza ion
Based P o ocol o Rou ing in Cloud
Ne wo k,” In . J. Compu . Ne wo ks Appl., ol.
8, no. 6, pp. 795–803, 2021, doi:
10.22247/ijcna/2021/210727.
[53] L. Mani, S. A umugam, and R. Jagana han,
“Pe o mance Enhancemen o Wi eless Senso
Ne wo k Using Feis y Pa icle Swa m
Op imiza ion P o ocol,” ACM In . Con .
P oceeding Se ., pp. 1–5, Dec. 2022, doi:
10.1145/3590837.3590907.
[54] M. H. Ja id, “Melanoma Skin Cance Da ase
o 10000 Images.” Kaggle, 2022. doi:
10.34740/KAGGLE/DSV/3376422.