E genç, Cansu; Ak aş, Ra e
A icle
Sec o -speci ic inancial o ecas ing wi h machine lea ning
algo i hm and SHAP in e ac ion alues
Financial In e ne Qua e ly
P o ided in Coope a ion wi h:
Uni e si y o In o ma ion Technology and Managemen , Rzeszów
Sugges ed Ci a ion: E genç, Cansu; Ak aş, Ra e (2025) : Sec o -speci ic inancial o ecas ing wi h
machine lea ning algo i hm and SHAP in e ac ion alues, Financial In e ne Qua e ly, ISSN
2719-3454, Sciendo, Wa saw, Vol. 21, Iss. 1, pp. 42-66,
h ps://doi.o g/10.2478/ iq -2025-0004
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/329895
S anda d-Nu zungsbedingungen:
Die Dokumen e au EconS o dü en zu eigenen wissenscha lichen
Zwecken und zum P i a geb auch gespeiche und kopie we den.
Sie dü en die Dokumen e nich ü ö en liche ode komme zielle
Zwecke e iel äl igen, ö en lich auss ellen, ö en lich zugänglich
machen, e eiben ode ande wei ig nu zen.
So e n die Ve asse die Dokumen e un e Open-Con en -Lizenzen
(insbesonde e CC-Lizenzen) zu Ve ügung ges ell haben soll en,
gel en abweichend on diesen Nu zungsbedingungen die in de do
genann en Lizenz gewäh en Nu zungs ech e.
Te ms o use:
Documen s in EconS o may be sa ed and copied o you pe sonal
and schola ly pu poses.
You a e no o copy documen s o public o comme cial pu poses, o
exhibi he documen s publicly, o make hem publicly a ailable on he
in e ne , o o dis ibu e o o he wise use he documen s in public.
I he documen s ha e been made a ailable unde an Open Con en
Licence (especially C ea i e Commons Licences), you may exe cise
u he usage igh s as speci ied in he indica ed licence.
h ps://c ea i ecommons.o g/licenses/by-nc-nd/3.0/
10.2478/ iq -2025-0004
Abs ac This s udy examines he applica ion o machine lea ning models o p edic inancial pe o mance
in a ious sec o s, using da a om 21 companies lis ed in he BIST100 index (2013-2023). The
p ima y objec i e is o assess he po en ial o hese models in imp o ing inancial o ecas accu-
acy and o emphasize he need o anspa en , explainable app oaches in inance. A ange o
machine lea ning models, including Linea Reg ession, Ridge, Lasso, Decision T ee, Bagging, Ran-
dom Fo es , AdaBoos , G adien Boos ing (GBM), Ligh GBM, and XGBoos , we e e alua ed. G a-
dien Boos ing eme ged as he bes -pe o ming model, wi h ensemble me hods gene ally
demons a ing supe io accu acy and s abili y compa ed o linea models. To enhance in e p e -
abili y, SHAP (SHapley Addi i e exPlana ions) alues we e u ilized, iden i ying he mos in luen-
ial a iables a ec ing p edic ions and p o iding insigh s in o model beha io . Sec o -based anal-
yses u he e ealed di e ences in model pe o mance and ea u e impac s, o e ing a g anula
unde s anding o inancial dynamics ac oss indus ies. The indings highligh he e ec i eness o
machine lea ning, pa icula ly ensemble me hods, in o ecas ing inancial pe o mance. The
s udy unde sco es he impo ance o using explainable models in inance o build us and sup-
po decision-making. By in eg a ing ad anced echniques wi h in e p e abili y ools, his e-
sea ch con ibu es o inancial echnology, ad ancing he adop ion o machine lea ning in da a-
d i en in es men s a egies.
JEL classi ica ion: C51, C52, C53
Keywo ds: Machine Lea ning Models, SHAP, Financial Fo ecas ing
Recei ed: 31.07.2024 Accep ed: 15.11.2024
Ci e his:
E genç, C. & Ak aş, R. (2025). Sec o -speci ic inancial o ecas ing wi h machine lea ning algo i hm and SHAP in e ac ion alues. Financial In e ne
Qua e ly 21(1), pp. 42-66.
© 2025 Cansu E genç and Ra e Ak aş, published by Sciendo. This wo k is licensed unde he C ea i e Commons A ibu ion-NonComme cial-
NoDe i a i es 3.0 License.
1 Anka a Yildi im Beyazi Uni e si y, Anka a, Tu key, e-mail: cansue [email protected]. , ORCID: h ps://o cid.o g/0000-0002-4722-0911.
2 Anka a Yildi im Beyazi Uni e si y, Anka a, Tu key, e-mail: ak [email protected]. , ORCID: h ps://o cid.o g/0009-0008-8033-4604.
Cansu E genç, Ra e Ak aş
Sec o -speci ic inancial o ecas ing wi h machine lea ning algo i hm and SHAP
in e ac ion alues
Financial In e ne Qua e ly 2025, ol. 21 / no. 1
BIST100 index is an impo an indica o o he Tu k-
ish s ock ma ke , which includes a wide ange o sec-
o s. The e o e, i p o ides a e y comp ehensi e da-
ase in e ms o inancial da a es ima ion, whe e he
pe o mance o machine lea ning models is e alua ed.
This esea ch e alua es he pe o mance o machine
lea ning models in inancial da a. This e alua ion is ca -
ied ou bo h on he gene al pe o mance o compa-
nies lis ed in BIST100 be ween 2013-2023 and on
a sec o al basis. As a esul , bo h he pe o mance o
machine lea ning models in inancial ma ke s and hei
pe o mance on a sec o al basis a e examined.
Ou s udy uses SHAP (SHapley Addi i e Explana-
ions) alues in addi ion o adi ional pe o mance
measu es o in e p e machine lea ning models. SHAP
alues inc ease he anspa ency and explainabili y o
complex ML algo i hms by p o iding insigh s in o ea-
u e impo ance and in e ac ion e ec s (Bha acha ya,
2022; Li, 2022; Bap is a e al., 2022; Bap is a, 2022). By
examining SHAP alues, his esea ch no only e alu-
a es he p edic i e accu acy o he models, bu also
cla i ies he key ac o s a ec ing inancial esul s. Thus,
he impo ance o he a iables in he models used o
he model is also examined. The esul s o his s udy
will be aluable o bo h academic esea ch and eal-
wo ld use and will p o ide impo an insigh s o in es-
o s, inancial analys s, and policy make s. The es o
his pape is s uc u ed as ollows. Sec ion 2 p o ides
a li e a u e e iew o he machine lea ning o ecas ing
models and ac o s ha in luence inancial pe o -
mance. Sec ion 3 p esen s he me hodology and sum-
ma izes nine machine lea ning models used o o ecas
inancial pe o mance. The esul s ob ained a e dis-
cussed in Sec ion 4. Finally, he conclusion is p esen ed
in Sec ion 5.
In ecen yea s, he e has been conside able p o-
g ess in inancial o ecas ing using machine lea ning
algo i hms. Machine lea ning models a e inc easingly
used in he inancial sec o o p edic s ock p ices and
classi ica ion (Sonka de, 2023). T adi ional models such
as linea eg ession a e s ill used (Gza e al., 2022).
Especially in p edic ing esul s based on inpu ea u es,
linea eg ession is a highly p e e ed model due o i s
simplici y and in e p e abili y (Rosenbusch e al., 2019;
Ryll e Seidens, 2019; Seno, 2023). Machine lea ning
models ha e been used in a wide ange o inancial
domains o pu poses such as c edi de aul p edic ion
and ou ism demand o ecas ing, p o iding aluable
insigh s o economic analysis and c isis de ec ion, and
ha e demons a ed he e sa ili y and e ec i eness o
hese algo i hms in di e en sec o s (Fan, 2023; Cla -
e ía e al., 2015; A een, 2020).
Financial pe o mance has always been c ucial o
companies, impac ing na ions globally. I is c ucial o
all coun ies and companies (Pe ini e al., 2011;
Ba auskai e & S eimikiene, 2020). In ecen yea s, he
combina ion o inance and a i icial in elligence has
no jus led o p og ess, bu a ans o ma ion in inan-
cial o ecas ing (Lin, 2019; Nguyen e al., 2022; A ela
& Jo dão, 2024). Machine lea ning algo i hms also play
a majo ole in his ans o ma ion. Because machine
lea ning algo i hms ha e p o ided ad anced ech-
niques ha can p ocess la ge amoun s o da a, iden i y
pa e ns, and make p edic ions wi h unp eceden ed
accu acy (Zhou e al., 2017; Mahalakshmi e al., 2022;
Bouche y & De Souza, 2020). Lea ning om his o ical
da a and adap ing o new in o ma ion, which is a ea-
u e o machine lea ning models, and he pe o mance
o models ha imp o e o e ime a e e y impo an
de elopmen s o inance (Pandey & Se gee a, 2022;
Ionescu & Diaconi a, 2023; Geo ge, 2024).
The place o accu a e inancial o ecas ing o i-
nancial ma ke s is undeniable (Penman, 2002; Samo-
nas, 2015; Kuma , 2017; Ba nhize & Ba nhize , 2019;
Sas y, 2020; Massei, 2023). In es o s educe hei i-
nancial isks and make in o med in es men s by mak-
ing he igh in es men decisions o accu a e inancial
o ecas s. Financial analys s, on he o he hand, make
ecommenda ions o ma ke pa icipan s in line wi h
he esul s ob ained om inancial o ecas s (Ramna h
e al., 2008; Samonas, 2015; Magnan e al., 2015).
Policy make s use inancial o ecas s o p e en
possible inancial c ises and guide he cu en econo-
my. Manage s can bene i om hese inancial o e-
cas s in hei s a egic decisions ega ding budge ing
(Ramna h e al., 2008; Oli a & Wa son, 2009; Magnan
e al., 2015; Ballings e al., 2015; Geng e al., 2015).
Wi h such esul s, machine lea ning models a e apidly
gaining accep ance in he ield o inance.
When machine lea ning models used in inancial
o ecas ing a e examined, i is seen ha me hods such
as neu al ne wo ks, decision ees and ensemble me h-
ods a e used. Each me hod has i s own ad an ages and
disad an ages (Ka al e al., 2013; P o os & Fawce ,
2013; Chen & Zhang, 2014; Naja abadi e al., 2015). The
pe o mances o hese me hods a y depending on he
s uc u e and size o he da a used. The ac ha hese
models gi e good esul s despi e he complex s uc u e
o inancial da a has caused hem o be p e e ed in
a eas such as c edi isks, s ock income, and es ima ing
he o al income o companies. In addi ion, he use o
big da a echnologies has enabled he p ocessing and
analysis o la ge da a se s, which has inc eased he p e-
cision and eliabili y o inancial o ecas s (Oli a & Wa -
son, 2009; P o os & Fawce , 2013; Chen & Zhang,
2014; Zhou e al., 2017).
Cansu E genç, Ra e Ak aş
Sec o -speci ic inancial o ecas ing wi h machine lea ning algo i hm and SHAP
in e ac ion alues
Financial In e ne Qua e ly 2025, ol. 21 / no. 1
in heal hca e (Deng e al., 2022) and ce ical cance
sc eening (Eda e anu e-Ibeh, 2024), line loss p edic ion
(Wang e al., 2017), A c ic na iga ion isk assessmen
(Yao e al., 2023), PM2.5 concen a ion es ima ion
(Pan, 2018) and pe mea e lux p edic ion in osmosis
p ocesses (Shi e al., 2022).
SHAP in e ac ion alues a e e y impo an o in-
c easing he accu acy o machine lea ning models. They
imp o e model in e p e abili y by cap u ing local in e -
ac ion e ec s be ween ea u es, especially in models
buil on inancial da a (O sini e al., 2022; Ze n e al.,
2023). In addi ion, SHAP in e ac ion alues ensu e con-
sis en indi idualized ea u e a ibu ion o ee com-
muni ies, p o iding consis en explana ions o in e ac-
ion e ec s in indi idual p edic ions (Lundbe g e al.,
2018; Mi chell e al., 2022). Using SHAP in e ac ion
alues makes models mo e unde s andable and allows
o a quan i a i e s udy o in e ac ion e ec s (Long e
al., 2022; Ma ini e al., 2022). As a esul , i p o ides
a uni ied app oach o in e p e complex model p edic-
ions and con ibu es o a mo e comp ehensi e unde -
s anding o model beha io (Li e al., 2020; Lundbe g e
al., 2020).
In his sec ion, we p esen he app oach used o
o ecas he inancial pe o mance o companies lis ed
on he BIST100 index om 2013 o 2023. The da ase
consis s o inancial me ics such as Ne Income, To al
Asse s, To al Liabili ies, and Sha eholde s' Equi y, which
se e as he independen a iables, while To al Re e-
nue is he a ge a iable. The da a is spli in o a ain-
ing se (80%) and a es se (20%) o ensu e p ope
e alua ion o model pe o mance. We employ en ma-
chine lea ning models: Linea Reg ession, Ridge Reg es-
sion, Lasso Reg ession, Decision T ee, Bagging, Random
Fo es , AdaBoos , G adien Boos ing (GBM), Ligh GBM,
and XGBoos . These models a e chosen due o hei
a ying complexi y and abili y o handle di e en ypes
o inancial da a. We apply se e al e alua ion me ics,
including Mean Squa ed E o (MSE), Roo Mean
Squa ed E o (RMSE), Mean Absolu e E o (MAE), and
Mean Absolu e Pe cen age E o (MAPE), o assess he
accu acy and obus ness o he models. Each model’s
p edic i e pe o mance is compa ed agains he es
se o e alua e i s abili y o gene alize.
To enhance model in e p e abili y, we use SHAP
(SHapley Addi i e exPlana ions) alues, which allow us
o assess he con ibu ion o each inpu a iable o he
model’s p edic ions. This helps in unde s anding he
impo ance o inancial me ics like Ne P o i , Long-
Te m Liabili ies, and To al Asse s in d i ing inancial
pe o mance ou comes. Addi ionally, we ensu e ha all
models a e con igu ed o accoun o he empo al na-
Linea eg ession is o en complemen ed by o he
algo i hms such as idge eg ession, lasso eg ession
and suppo ec o eg ession o inc ease p edic ion
accu acy (Xiao e al., 2020; Yoo e al., 2022). In addi-
ion, s udies compa e pe o mance wi h models such
as Random Fo es , XGBoos and LSTM (Sonka de,
2023). Wi h he use o machine lea ning models in he
inancial sec o , which model will i he da a be e has
become an impo an issue (Long e al., 2022; Akin ino-
la, 2024). Decision ees, which a e a equen ly used
model among machine lea ning models, a e p e e ed
due o hei e ec i eness, in e p e abili y and ease o
isualiza ion (Kou ellis e al., 2016; Moshko , 1997;
Azad e al., 2022; Pooji ha & Kanagasabai, 2022). The
s uc u e o inancial da a is complex and a iable, and
G adien Boos ing, which has shown signi ican success
in a ious p ac ical applica ions due o i s abili y o han-
dle complex ela ionships and p oduce accu a e p edic-
ions in he use o such da a, can be p e e ed (Na ekin
& Knoll, 2013; Chen, 2016; Kadiyala & Kuma , 2018;
Da is e al., 2020). Along wi h his me hod, adian
Boos ing algo i hms such as XGBoos , Ligh GBM and
o he s ha e become popula choices in he machine
lea ning communi y due o hei e ec i eness in im-
p o ing model pe o mance and p edic ion accu acy
(Mienye & Sun, 2022; Si ingo ingo e al., 2021; Zhang e
al., 2011).
Ligh GBM has been compa ed wi h o he machine
lea ning models such as Random Fo es , XGBoos , and
adi ional g adien boos ing in he li e a u e, and has
ou pe o med hese models in e ms o pe o mance,
speed, accu acy, and e iciency (F az, 2024; G issa e
al., 2020; Unal e al., 2021; Jiang, 2024). Ligh GBM has
been success ully used in a ious ields, including
heal h, en i onmen al science, inance, and geology
(Ru o e al., 2021; Su e al., 2021; Pa k e al., 2021;
Dong e al., 2022; Ko e al., 2022; Jiang, 2024; Xiang,
2024; Wang, 2024). Fu he mo e, he e sa ili y o
Ligh GBM is e iden in i s applica ions in a ious ields
such as aul de ec ion in wind u bines (Tang e al.,
2020), in usion de ec ion in IoT sys ems (Zhao e al.,
2023), aud de ec ion in banking da a (Hashemi e al.,
2023), and malwa e de ec ion (Onoja e al., 2022). An-
o he al e na i e o Ligh GBM is he XGBoos model.
The XGBoos algo i hm has been shown o exhibi e y
high accu acy and pe o mance on a ious da ase s
(Chen, 2016; Ka eem e al., 2023). I has been success-
ully used in a ious ields including elec ion p edic ion
(Suacana, 2024), ai c a icing se e i y assessmen (Li e
al., 2020), su ace milling accu acy (Abbas, 2023), anal-
ysis o imbalanced da a (Zhang e al., 2022), and p edic-
ion and op imiza ion asks (Zhang, 2024). I has been
used in a ious applica ions such as jaundice de ec ion
in newbo ns (Abdul azzak, 2024), aul de ec ion in
pho o ol aic panels (Sai am, 2020), ou come p edic ion
Cansu E genç, Ra e Ak aş
Sec o -speci ic inancial o ecas ing wi h machine lea ning algo i hm and SHAP
in e ac ion alues
Financial In e ne Qua e ly 2025, ol. 21 / no. 1
en ial ou comes, aiding in unde s anding complex sce-
na ios and making p edic ions based on inpu da a (Lo
e al., 2014). The p edic ion o a decision ee is gi en
by:
(4)
whe e M is he numbe o e minal nodes, wm is he
p edic ed alue in egion Rm, and I(·) is an indica o
unc ion.
Bagging, sho o boo s ap agg ega ing, is a ech-
nique ha in ol es gene a ing mul iple e sions o
a p edic o by esampling he aining da a and hen
agg ega ing hese p edic o s o c ea e a mo e obus
and accu a e model (B eiman, 1996; Gianola e al.,
2014; Solo e al., 2023). Bagging p edic ion is:
(5)
whe e B is he numbe o boo s ap samples and b(x) is
he p edic ion om he b - h boo s ap sample.
Random Fo es is an ensemble supe ised lea ning
algo i hm known o i s high accu acy in classi ica ion
asks (Ilma e al., 2023; Sandhya & Padyana, 2021;
Genue , 2012). I gene a es mul iple decision ees by
esampling he da a and agg ega ing he p edic ions,
esul ing in a obus and accu a e model. (Genue ,
2012, S obl e al., 2008; Mishina e al., 2015; Kulka ni
& Sinha, 2012). Random Fo es p edic ion is:
(6)
whe e T is he numbe o ees, and (x) is he p edic-
ion o he $ $ - h ee.
AdaBoos , sho o Adap i e Boos ing, is an en-
semble lea ning me hod ha combines mul iple weak
lea ne s o c ea e a s ong classi ie (Paul e al., 2009;
Mei & Rä sch, 2003). I i e a i ely adjus s he weigh s
o inco ec ly classi ied ins ances o ocus on di icul
cases, imp o ing he o e all model pe o mance (Wang
e al., 2022; Yin e al., 2017; Si e al., 2022). AdaBoos
p edic ion is:
(7)
whe e T is he numbe o ees, α is he weigh as-
signed o he $ $ - h ee based on i s accu acy, and
+(x) is he p edic ion o he - h ee.
u e o he da a, al hough no explici ime-se ies mod-
els we e used. Neighbo ing ec o s o da a a e consid-
e ed wi hin he amewo k o machine lea ning models
o ensu e ha he ime con ex is espec ed du ing
aining and p edic ions.
Linea eg ession analysis assumes a linea ela-
ionship among mul iple a iables (Sch oede e al.,
2016). The gene al Linea Reg ession model can be
s a ed by he equa ion below:
(1)
whe e, yi dependen a iable, xi explana o y a iables,
β0 cons an e m, βk slope coe icien s o each explan-
a o y a iable, Ɛi he model's e o e m.
Ridge eg ession is an ex ension o linea eg es-
sion, known o i s bias- a iance ade-o con ol ha
p o ides a balance be ween model complexi y and
gene aliza ion pe o mance, is a aluable echnique
used o add ess mul icollinea i y in da ase s whe e
independen a iables a e highly co ela ed
(Mal house, 1999; Kib ia & Saleh, 2004; Khala , 2012;
Kuma e al., 2021). By adding a penal y e m o he
OLS me hod, idge eg ession helps o s abilize he p e-
dic ions and p e en o e i ing, making i a mo e elia-
ble and consis en me hod o modeling ela ionships
be ween a iables (Xin & Khalid, 2018; Wei & Diğe le i,
2020; Li, 2024). Ridge eg ession minimizes he ollow-
ing cos unc ion:
(2)
whe e λ is he egula iza ion pa ame e .
Lasso eg ession is a widely used echnique in e-
g ession analysis known o i s abili y o pe o m a ia-
ble selec ion and egula iza ion (Raja a nam e al.,
2015; Signo ino & Ki chne , 2018; F iedman e al.,
2010). Lasso eg ession minimizes he ollowing cos
unc ion:
(3)
whe e λ is he egula iza ion pa ame e .
A decision ee is a decision suppo ool ha uses
a ee-like g aph o ep esen decisions and hei po-
0 1 1 2 2 ...
i i i k ki i
y X X X
= + + + + +
22
0
1 1 1
()
pp
n
i j ij j
i j j
a gmin y x
= = =
= − − +
2
0
1 1 1
()
pp
n
i j ij j
i j j
a gmin y x
= = =
= − − +
1
( ) ( )
M
mm
m
x w I x R
=
=
1
1
( ) ( )
B
b
b
x x
B=
=
1
1
( ) ( )
T
b
x x
T=
=
1
( ) ( )
T
x x
=
=
Cansu E genç, Ra e Ak aş
Sec o -speci ic inancial o ecas ing wi h machine lea ning algo i hm and SHAP
in e ac ion alues
Financial In e ne Qua e ly 2025, ol. 21 / no. 1
(11)
(12)
(13)
(14)
(15)
SHapley Addi i e exPlana ions (SHAP) alues a e
a me hod oo ed in coope a i e game heo y ha aims
o p o ide a ai alloca ion o impo ance alues o ea-
u es in machine lea ning models (Uddin e al., 2022).
This app oach has been u ilized in a ious s udies o
enhance he in e p e abili y and anspa ency o ma-
chine lea ning models ac oss di e en domains. Fo
ins ance, SHAP alues ha e been employed o in e p e
he ou pu s o suppo ec o machines, andom o -
es s, con olu ional neu al ne wo ks, and long sho -
e m memo y models in o ecas ing clima ic wa e bal-
ance (Uddin e al., 2022). Addi ionally, SHAP has been
used o in e p e models in p edic ing sepsis in-hospi al
mo ali y (Zhang, 2024), au oma ing da a cen e ope a-
ions (Geb eyesus, 2024), and de eloping p ognos ic
models o c i ically ill pa ien s (Fan e al., 2023). The
applica ion o SHAP alues ex ends o di e se a eas
such as p edic ing opical cyclogenesis (Loi, 2024),
e alua ing hospi al mobili y (San ama o, 2024), p e-
dic ing gou associa ed wi h die a y ac o s (Cao, 2024),
and op imizing pho odeg ada ion a e p edic ions
(Schossle , 2023). By le e aging SHAP alues, esea ch-
e s ha e gained deepe insigh s in o model p edic ions,
ea u e impo ance, and he speci ic con ibu ions o
a iables o he ou comes o machine lea ning models
(Cao, 2024). Fu he mo e, SHAP alues ha e been in-
s umen al in enhancing he in e p e abili y, explaina-
bili y, and ai ness o machine lea ning models (Hickey
e al., 2020).
Fo a model and inpu ea u es , he SHAP alue
o a ea u e is gi en by:
(16)
G adien Boos ing is a powe ul ensemble machine
lea ning echnique ha i e a i ely builds a se ies o
weak lea ne s, ypically decision ees, o c ea e
a s ong p edic i e model. By ocusing on he e o s o
he p e ious models du ing aining, G adien Boos ing
aims o imp o e p edic ion accu acy by minimizing he
o e all loss unc ion (Zhang e al., 2011; May
& Schmid, 2014; Johnson & Zhang, 2014). G adien
Boos ing p edic ion is:
(8)
whe e T is he numbe o ees, is he lea ning a e,
and (x) is he - h ee ained o p edic he esiduals
o he p e ious ees.
Ligh GBM, sho o Ligh G adien Boos ing Ma-
chine, is an ex emely e icien algo i hm designed o
g adien -boos ing decision ees (Jiang, 2024).
Ligh GBM p edic ion is:
(9)
whe e T is he numbe o ees, and (x) is he p edic-
ion o he - h ee using he Ligh GBM amewo k,
which employs g adien -based one-side sampling and
exclusi e ea u e bundling.
XGBoos , sho o Ex eme G adien Boos ing, is
a powe ul machine lea ning algo i hm enowned o
i s scalabili y, speed, and accu acy (Chen, 2016).
XGBoos p edic ion is:
(10)
whe e T is he numbe o ees, and (x) is he p edic-
ion o he - h ee using he XGBoos algo i hm.
which op imizes o a equla ized obiec i e o p e en
o e i ing.
The e alua ion o hese models was conduc ed
using se e al key pe o mance me ics: Mean Squa ed
E o (MSE), Roo Mean Squa ed E o (RMSE), Mean
Absolu e E o (MAE), Mean Absolu e Pe cen age E o
(MAPE), and ela i e Roo Mean Squa ed E o
( RMSE). The e alua ion o he machine lea ning mod-
els in his s udy is based on se e al key pe o mance
me ics ha quan i y he accu acy and obus ness o
he p edic ions. The me ics used a e as ollows:
1
( ) ( )
T
x x
=
=
1
( ) ( )
T
x x
=
=
1
( ) ( )
T
x x
=
=
2
1()
ii
MSE Y Y
n
=−
2
1()
ii
RMSE Y Y
n
=−
1
i
MAE Y Y
n
=−
1
100/ % nii
ii
yy
MAPE ny
=
−
=
2
1
1()
n
i i i
yy
n
RMSE y
=−
=
!( 1)![ ( {}) ( )]
!s s s s
SN
S N S
i x i x
N
−−
= −
Cansu E genç, Ra e Ak aş
Sec o -speci ic inancial o ecas ing wi h machine lea ning algo i hm and SHAP
in e ac ion alues
Financial In e ne Qua e ly 2025, ol. 21 / no. 1
In his a icle, we in es iga e he impac o a ia-
bles a ec ing inancial pe o mance on o al e enue.
The da a consis s o 21 companies lis ed in BIST100 o
a 10-yea pe iod (2013-2023). Figu e 1 shows he sec-
o al dis ibu ion o companies.
Whe e, N is he se o all ea u es, S is any subse o N
ha does no include ea u e I, s(xs U {i}) is he p edic-
ion o he model when ea u e i is included in subse S,
s(xs) is he p edic ion o he model when ea u e i is
no included.
Figu e 1: Sec o al Dis ibu ion o Companies
Sou ce: Au ho ’s own wo k.
ou s udy, Ne Income, To al Asse s, To al Liabili ies,
and Sha eholde s’ Equi y, Sho - e m Liabili ies, Long-
e m Liabili ies we e ea ed as independen ea u es,
while To al Re enue se ed as he ou pu o a ge
ea u e. Figu e 2 shows he ROA o each company
om 2013 o 2023.
This s udy is sepa a ed in o aining (80%) and
es ing (20%) da ase s o compa e he pe o mance o
a ious machine lea ning models. The da ase is an-
domly spli , wi h 80% used as he aining da ase and
he emaining 20% as he es ing se . This app oach is
commonly used in p io s udies (Abellán & Man as,
2014; An unes e al., 2017; Ben Jabeu e al., 2020). In
Figu e 2: ROA o each company om 2013 o 2023
Sou ce: Au ho ’s own wo k.
Au omo i e and
Au omo i e Sub-
Indus y
Ene gy
Food and
Be e ages
Holding and
In es men s
S eel
Re ail
Telecommunica ions
Cemen and
Cons uc ion
Ma e ials
Chemicals and
Sma Ma e ials
Home Appliances
and Elec onics
I on
Re ail
Cansu E genç, Ra e Ak aş
Sec o -speci ic inancial o ecas ing wi h machine lea ning algo i hm and SHAP
in e ac ion alues
Financial In e ne Qua e ly 2025, ol. 21 / no. 1
(MAPE), and ela i e Roo Mean Squa ed E o
( RMSE). Table 1 summa izes he pe o mance me ics
o each model. Appendix 1 shows he pe o mance o
machine lea ning models o e he es sample.
The pe o mance o a ious machine lea ning mod-
els was e alua ed using i e key me ics: Mean Squa ed
E o (MSE), Roo Mean Squa ed E o (RMSE), Mean
Absolu e E o (MAE), Mean Absolu e Pe cen age E o
Table 1: MSE, RMSE, MAE, MAPE and RMSE alues
Model MSE RMSE MAE MAPE RMSE
Linea eg ession 0.0040 0.0635 0.0430 41.22% 0.736
Ridge eg ession 0.0040 0.0635 0.0430 41.22% 0.736
Lasso Reg ession 0.0040 0.0635 0.0430 41.22% 0.736
Decision T ees 0.0046 0.0676 0.0456 118.58% 0.783
Bagging 0.0016 0.0399 0.0239 4.19% 0.462
Random Fo es s 0.0016 0.0403 0.0246 4.55% 0.466
Adaboos 0.0018 0.0428 0.0269 2.83% 0.496
GBM 0.0014 0.0378 0.0235 13.92% 0.438
Ligh GBM 0.0044 0.0695 0.0719 69.84% 0.777
XGBoos 0.0046 0.0679 0.0616 73.54% 0.786
Sou ce: Au ho ’s own wo k.
ion accu acy by ocusing on misclassi ied ins ances.
AdaBoos 's i e a i e app oach o adjus ing he weigh s
o misclassi ied ins ances con ibu es o i s enhanced
pe o mance and eliabili y. G adien Boos ing (GBM)
ou pe o ms mos models wi h he lowes MSE o
0.0014 and RMSE o 0.0378. The model's MAE and
MAPE alues indica e high accu acy and p ecision in
p edic ions, making i a obus choice o inancial o e-
cas ing. GBM's abili y o i e a i ely imp o e upon e -
o s made by p e ious models esul s in supe io p e-
dic i e capabili ies and obus ness. Also, Ligh GBM,
known o i s e iciency, shows highe e o me ics
compa ed o o he boos ing me hods. This could be
due o he model's sensi i i y o he da ase cha ac e -
is ics o he hype pa ame e se ings used in his s udy.
I s MAPE o 69.84% indica es conside able p edic ion
e o s in ce ain ins ances, sugges ing ha u he un-
ing and adjus men may be needed o op imize i s pe -
o mance o his speci ic da ase . Simila ly, XGBoos ,
ano he popula boos ing algo i hm, pe o ms akin o
Decision T ees, wi h an MSE o 0.0046 and an RMSE o
0.0679. Howe e , i shows a ela i ely high MAPE o
73.54%, indica ing ha i may no be he bes i o
his speci ic da ase wi hou u he uning. The highe
e o me ics sugges ha XGBoos 's de aul se ings
migh no be ully op imized o he inancial o e-
cas ing ask a hand. The a ia ion in MAPE ac oss
hese models can be a ibu ed o hei espec i e abili-
ies o cap u e complex ela ionships in he inancial
da a. Simple models like Linea Reg ession, Ridge, and
Lasso s uggle wi h hese in icacies, leading o highe
e o a es. On he con a y, ensemble me hods like
Bagging, Random Fo es s, and G adien Boos ing end
o mi iga e o e i ing and handle complex da a ela-
The linea models, including Linea Reg ession,
Ridge Reg ession, and Lasso Reg ession, exhibi iden i-
cal pe o mance me ics. These models a e cha ac e -
ized by hei simplici y and baseline na u e, which is
e lec ed in he ela i ely high alues o MSE, RMSE,
MAE, and RMSE. The MAPE o hese models is no a-
bly la ge a 41.22%, indica ing ha hey may s uggle o
cap u e he complex ela ionships wi hin he inancial
da a e ec i ely. This limi a ion sugges s ha while
hese models a e s aigh o wa d o in e p e , hey a e
no well-sui ed o accu a ely p edic ing inancial pe -
o mance in his con ex . The Decision T ee model
shows a highe MSE and RMSE compa ed o he linea
models, wi h an excep ionally high MAPE o 118.58%.
This high MAPE sugges s ha he Decision T ee model
ends o o e i he da a, making i less eliable o p e-
dic ion despi e i s in e p e abili y. The o e i ing is
likely due o he model's endency o cap u e noise and
luc ua ions in he aining da a, leading o poo gene -
aliza ion o new da a. On he o he hand, ensemble
me hods such as Bagging and Random Fo es s demon-
s a e signi ican ly be e pe o mance han he indi id-
ual Decision T ee model. These models exhibi lowe
MSE, RMSE, and MAE alues, wi h Bagging showing
a sligh ly be e pe o mance han Random Fo es s.
The MAPE alues o Bagging and Random Fo es s a e
imp essi ely low a 4.19% and 4.55%, espec i ely, indi-
ca ing s ong p edic i e accu acy and s abili y. These
esul s highligh he e ec i eness o ensemble me h-
ods in educing a iance and imp o ing he obus ness
o p edic ions. AdaBoos pe o ms well, wi h an MSE o
0.0018 and an RMSE o 0.0428. The model shows
a ema kably low MAPE o 2.83%, unde sco ing i s abil-
i y o handle complex da a and imp o e o e all p edic-
Cansu E genç, Ra e Ak aş
Sec o -speci ic inancial o ecas ing wi h machine lea ning algo i hm and SHAP
in e ac ion alues
Financial In e ne Qua e ly 2025, ol. 21 / no. 1
s and how each model pe o med wi hin speci ic indus-
ies. Table 2 p esen s he Mean Squa ed E o (MSE)
alues o each sec o and model combina ion. Figu e
4 shows Compa ison o MSE Values Ac oss Di e en
Sec o s and Models.
ionships mo e e ec i ely, esul ing in lowe MAPE and
be e p edic i e pe o mance.
The pe o mance o he machine lea ning models
was u he analyzed ac oss di e en sec o s o unde -
Table 2: MSE alues o sec o
Sec o s
Linea
eg ession Ridge Lasso DT Bagging
MSE
Food and Be e age 0.00266 0.00266 0.00266 0.00148 0.00022
Cemen and Cons uc ion Ma e ials 0.00203 0.00203 0.00203 0.00197 0.00121
Chemis y and Sma Ma e ials 0.00307 0.00307 0.00307 0.00061 0.00024
Ene gy 0.00423 0.00423 0.00423 0.00127 0.00035
Home Appliances and Elec onics 0.00102 0.00102 0.00102 0.00000 0.00004
Au omo i e and Au omo i e Sub-Indus y 0.00805 0.00805 0.00805 0.00073 0.00070
Holding and In es men 0.00137 0.00137 0.00137 0.00062 0.00055
I on-S eel 0.00172 0.00172 0.00172 0.00034 0.00046
Re ail 0.00861 0.00861 0.00861 0.00045 0.00012
Telecommunica ions 0.00231 0.00231 0.00231 0.00063 0.00044
Sec o s RF Adaboos GBM Ligh GBM XGBoos
MSE
Food and Be e age 0.00021 0.00004 0.00017 0.00455 0.00266
Cemen and Cons uc ion Ma e ials 0.00125 0.00154 0.00133 0.00355 0.00203
Chemis y and Sma Ma e ials 0.00027 0.00019 0.00008 0.00357 0.00307
Ene gy 0.00033 0.00031 0.00015 0.00658 0.00423
Home Appliances and Elec onics 0.00004 0.00002 0.00006 0.00500 0.00102
Au omo i e and Au omo i e Sub-Indus y 0.00070 0.00030 0.00040 0.00438 0.00805
Holding and In es men 0.00058 0.00054 0.00026 0.00576 0.00137
I on-S eel 0.00046 0.00002 0.00015 0.00451 0.00172
Re ail 0.00012 0.00009 0.00012 0.00710 0.00861
Telecommunica ions 0.00050 0.00069 0.00046 0.00320 0.00231
Sou ce: Au ho ’s own wo k.
bes pe o mance in his sec o wi h MSE alues o
0.00008 and 0.00024, espec i ely. These esul s high-
ligh he e ec i eness o hese ensemble me hods in
cap u ing he complexi y o da a in he chemis y and
sma ma e ials sec o . In he ene gy sec o , GBM
s ands ou wi h an MSE o 0.00015, ollowed by Bag-
ging wi h an MSE o 0.00035. These models demon-
s a e supe io p edic i e accu acy, sugges ing hey a e
well-sui ed o o ecas ing inancial pe o mance in he
ene gy indus y.
The Bagging model pe o ms excep ionally well in
his sec o , achie ing he lowes MSE o 0.00022. Ran-
dom Fo es ollows closely wi h an MSE o 0.00021,
indica ing s ong p edic i e pe o mance. AdaBoos
also pe o ms well wi h an MSE o 0.00004, sugges ing
high accu acy in his sec o . Bagging and Random Fo -
es s show be e pe o mance in his sec o compa ed
o o he models, wi h MSE alues o 0.00121 and
0.00125, espec i ely. GBM also pe o ms well wi h an
MSE o 0.00133, indica ing good p edic i e capabili ies
in his indus y. GBM and Bagging models exhibi he
Cansu E genç, Ra e Ak aş
Sec o -speci ic inancial o ecas ing wi h machine lea ning algo i hm and SHAP
in e ac ion alues Financial In e ne Qua e ly 2025, ol. 21 / no. 1
Moshko , M. (1997). Algo i hms o cons uc ing o decision ees. In: P inciples o Da a Mining and Knowledge Dis-
co e y: Fi s Eu opean Symposium (pp. 335-342). Sp inge , Be lin.
Naja abadi, M.M., Villanus e, F., Khoshgo aa , T.M., Seliya, N., Wald, R. & Muha emagic, E. (2015). Deep lea ning
applica ions and challenges in big da a analy ics. Jou nal o Big Da a, 2(1), 1-21.
Na ekin, A. & Knoll, A. (2013). G adien boos ing machines, a u o ial. F on ie s in Neu o obo ics, 7, 21.
Nguyen, D.K., Se mpinis, G. & S asinakis, C. (2023). Big da a, a i icial in elligence and machine lea ning: A ans o m-
a i e symbiosis in a o o inancial echnology. Eu opean Financial Managemen , 29(2), 517-548.
Oli a, R. & Wa son, N. (2009). Managing unc ional biases in o ganiza ional o ecas s: A case s udy o consensus
o ecas ing in supply chain planning. P oduc ion and Ope a ions Managemen , 18(2), 138-151.
Onoja, M., Jegede, A., Blamah, N., Abimbola, O.V. & Omo ehinwa, T.O. (2022). EEMDS: E icien and e ec i e mal-
wa e de ec ion sys em wi h hyb id model based on Xcep ionCNN and Ligh GBM algo i hm. Jou nal o Compu ing
and Social In o ma ics, 1(2), 42-57, h ps://doi.o g/10.33736/jcsi.4739.2022.
O sini, N., Moo e, A. & Wolk, A. (2022). In e ac ion analysis based on Shapley alues and ex eme g adien boos ing:
A ealis ic simula ion and applica ion o a la ge epidemiological p ospec i e s udy. F on ie s in Nu i ion, 9, 1-8.
Pan, B. (2018, Feb ua y). Applica ion o XGBoos algo i hm in hou ly PM2.5 concen a ion p edic ion. In: IOP Con e -
ence Se ies: Ea h and En i onmen al Science (Vol. 113, p. 012127). IOP Publishing, B is ol.
Pandey, M.K. & Se gee a, I. (2022). A i icial in elligence impac e alua ion: T ans o ming pa adigms in inancial in-
s i u ions. Mi Èkonomiki I Up a leniâ, 22(1), 147-164, h ps://doi.o g/10.25205/2542-0429-2022-22-1-147-164.
Pa k, S., Son, S., Bae, J., Lee, D., Kim, J.J. & Kim, J. (2021). Robus spa io empo al es ima ion o PM concen a ions
using boos ing-based ensemble models. Sus ainabili y, 13(24), 1-15.
Paul, B., A hi han, G. & Mu y, M.N. (2009). Speeding up AdaBoos classi ie wi h andom p ojec ion. In: 2009 Se -
en h In e na ional Con e ence on Ad ances in Pa e n Recogni ion (pp. 251-254), IEEE, New Je sey.
Penman, S. H. (2002). The quali y o inancial s a emen s: Pe spec i es om he ecen s ock ma ke bubble. Pape s
SSRN 319262, 1-44, h p://dx.doi.o g/10.2139/ss n.319262.
Pe mana, S., Rosadi, R. & Nikki, N. (2022). Applica ion o classi ica ion algo i hm o sales p edic ion. TEKNOKOM,
5(2), 119-124, h ps://doi.o g/10.31943/ eknokom. 5i2.77.
Pe ini, F., Russo, A., Tenca i, A. & Vu o, C. (2011). Decons uc ing he ela ionship be ween co po a e social and
inancial pe o mance. Jou nal o Business E hics, 102, 59-76, h ps://doi.o g/10.1007/s10551-011-1194-1.
Pooji haa, M. & Mala hib, K. (2022). Decision ee o e suppo ec o machine o be e accu acy in iden i ying he
p oblem based on he I is lowe . Ad ances in Pa allel Compu ing Algo i hms, Tools and Pa adigms, 41, 209-217.
P asad, A. & Bakhshi, P. (2022). Fo ecas ing he di ec ion o daily changes in he India VIX index using machine lea n-
ing. Jou nal o Risk and Financial Managemen , 15(12), 1-16, h ps://doi.o g/10.3390/j m15120552.
P o os , F. & Fawce , T. (2013). Da a science and i s ela ionship o big da a and da a-d i en decision making. Big
Da a, 1(1), 51-59, h ps://doi.o g/10.1089/big.2013.1508.
Raja a nam, B., Robe s, S., Spa ks, D. & Dalal, O. (2015). Lasso eg ession: Es ima ion and sh inkage ia he limi o
Gibbs sampling. Jou nal o he Royal S a is ical Socie y: Se ies B S a is ical Me hodology, 78(1), 153-174.
Ramna h, S., Rock, S. & Shane, P. (2008). The inancial analys o ecas ing li e a u e: A axonomy wi h sugges ions
o u he esea ch. In e na ional Jou nal o Fo ecas ing, 24(1), 34-75.
Roha gi, S., Singh, K.K. & Jasuja, D. (2021). Compa a i e analysis o machine lea ning algo i hm o o ecas Indian
s ock ma ke . In 2021 In e na ional Con e ence on Ad ance Compu ing and Inno a i e Technologies in Enginee -
ing (ICACITE) (pp. 278-283). IEEE, New Je sey, h ps://doi.o g/10.1109/ICACITE51222.2021.9404642.
Cansu E genç, Ra e Ak aş
Sec o -speci ic inancial o ecas ing wi h machine lea ning algo i hm and SHAP
in e ac ion alues Financial In e ne Qua e ly 2025, ol. 21 / no. 1
Rosenbusch, H., Soldne , F., E ans, A.M. & Zeelenbe g, M. (2021). Supe ised machine lea ning me hods in psycholo-
gy: A p ac ical in oduc ion wi h anno a ed R code. Social and Pe sonali y Psychology Compass, 15(2), 1-25.
Ru o, D., Debelee, T., Iben hal, A. & Nege a, W. (2021). Diagnosis o diabe es melli us using g adien boos ing ma-
chine (Ligh GBM). Diagnos ics, 11(9), 1-14, h ps://doi.o g/10.3390/diagnos ics11091714
Ryll, L. & Seidens, S. (2019). E alua ing he pe o mance o machine lea ning algo i hms in inancial ma ke o e-
cas ing: A comp ehensi e su ey, a Xi p ep in , 1906, h ps://doi.o g/10.48550/a Xi .1906.07786.
Sai am, S., S ini asan, S., Ma a io i, G., Suba h a, B., Ma hisen, G. & Beki oglu, K. (2020). Explainable Incipien Faul
De ec ion Sys ems o Pho o ol aic Panels. a Xi p ep in , 2011, h ps://doi.o g/10.48550/a Xi .2011.09843.
Samonas, M. (2015). Financial o ecas ing, analysis, and modelling: A amewo k o long- e m o ecas ing. John
Wiley & Sons, Hoboken.
Sandhya, V. & Padyana, A. (2021). Machine lea ning based c op yield p edic ion on geog aphical and clima ic da a.
In: 2021 Six h In e na ional Con e ence on Image In o ma ion P ocessing (ICIIP) (pp. 186-191). IEEE, New Je sey.
Sas y, V.V.L.N. (2020). A i icial in elligence in inancial se ices and banking indus y. Idea Publishing, London.
Shi, F., Lu, S., Gu, J., Lin, J., Zhao, C., You, X. & Lin, X. (2022). Modeling and e alua ion o he pe mea e lux in o wa d
osmosis p ocess wi h machine lea ning. Indus ial & Enginee ing Chemis y Resea ch, 61(49), 18045-18056.
Si, Z., Niu, H. & Wang, W. (2022). C edi Risk Assessmen by a Compa ison Applica ion o Two Boos ing Algo i hms.
In: Fuzzy Sys ems and Da a Mining VIII (pp. 34-40). IOS P ess, Ams e dam.
Signo ino, C. & Ki chne , A. (2018). Using lasso o model in e ac ions and nonlinea i ies in su ey da a. Su ey P ac-
ice, 11(1), 1-10, h ps://doi.o g/10.29115/SP-2018-0005.
Si ingo ingo, R., Pe angin, R. & Jamaluddin, J. (2021). Model hib id gene ic-XGBoos dan p incipal componen analy-
sis pada segmen asi dan pe amalan pasa . Me homika Ju nal Manajemen In o ma ika Dan Kompu e isasi
Akun ansi, 5(2), 97-103, h ps://doi.o g/10.46880/jmika.Vol5No2.pp97-103.
Solo , J. A., Ba be , R.F. & Wille , R. (2024). Bagging p o ides assump ion- ee s abili y. Jou nal o Machine Lea ning
Resea ch, 25(131), 1-35.
Sonka de, G. (2023). Fo ecas ing s ock ma ke p ices using machine lea ning and deep lea ning models: A sys ema ic
e iew, pe o mance analysis and discussion o implica ions. In e na ional Jou nal o Financial S udies, 11(3),
1-22, h ps://doi.o g/10.3390/ij s11030094.
S obl, C., Boules eix, A., Kneib, T., Augus in, T. & Zeileis, A. (2008). Condi ional a iable impo ance o andom o -
es s. BMC Bioin o ma ics, 9(1), 1-11.
Su, H., Lu, X., Chen, Z., Zhang, H., Lu, W. & Wu, W. (2021). Es ima ing coas al chlo ophyll-a concen a ion om ime-
se ies OLCI da a based on machine lea ning. Remo e Sensing, 13(4), 1-21.
Suacana, I. (2024). Op imizing he 2024 go e no elec ion quick coun wi h ex eme g adien boos ing (XGBoos ) o
inc ease o ing p edic ion accu acy. In e na ional Jou nal o So wa e Enginee ing and Compu e Science, 4(1),
91-106, h ps://doi.o g/10.35870/ijsecs. 4i1.2286.
Tang, M., Zhao, Q., Ding, S., Wu, H., Li, L., Wen, L. & Huang, B. (2020). An imp o ed Ligh GBM algo i hm o online
aul de ec ion o wind u bine gea boxes. Ene gies, 13(4), 807-823.
Uddin, M.N., Li, L.Z., Deng, B.Y. & Ye, J. (2023). In e p e able XGBoos –SHAP machine lea ning echnique o p edic
he comp essi e s eng h o en i onmen - iendly ice husk ash conc e e. Inno a i e In as uc u e Solu ions,
8(5), 147-168, h ps://doi.o g/10.1007/s41062-023-01122-9.
Cansu E genç, Ra e Ak aş
Sec o -speci ic inancial o ecas ing wi h machine lea ning algo i hm and SHAP
in e ac ion alues Financial In e ne Qua e ly 2025, ol. 21 / no. 1
Ünal, A. F., Kaleli, A. Y., Ummak, E. & Albay ak, Ö. (2021, Augus ). A Compa ison o S a e-o - he-A Machine Lea n-
ing Algo i hms on Faul Indica ion and Remaining Use ul Li e De e mina ion by Teleme y Da a. In: 8 h In e na-
ional Con e ence on Fu u e In e ne o Things and Cloud (pp. 79-85). IEEE, New Je sey.
Wang, L., Ke n, R., Yu, E., Choi, S. & Pan, J. (2023). In elliSleepSco e , a so wa e package wi h a g aphic use in e -
ace o au oma ed sleep s age sco ing in mice based on a ligh g adien boos ing machine algo i hm. Scien i ic
Repo s, 13(1), 1-11, h ps://doi.o g/10.1038/s41598-023-31288-2.
Wang, M. (2024). Iden i ica ion o mine wa e sou ce based on TPE-Ligh GBM. Scien i ic Repo s, 14(1), 1-11.
Wang, P., Xie, M., Wang, X., Yu, J., Chen, E., Zhou, Z., Niu, Y., Song, W., Ni, Q. & Zhu, J. (2022). Compa ison o nomo-
g am wi h machine lea ning echniques o p edic ion o o e all su i al in pa ien s wi h e ope i oneal liposa -
coma, Resea ch Squa e, 1, 1-20.
Wang, S., Peng ei, D. & Tian, Y. (2017). A no el me hod o s a is ical line loss es ima ion o dis ibu ion eede s
based on eede clus e and modi ied XGBoos . Ene gies, 10(12), 1-17.
Wei, C. (2024). Compa ison o di e en machine lea ning classi ica ion models o p edic ing deep ein h ombosis
in lowe ex emi y ac u es. Scien i ic Repo s, 14(1), 1-8.
Wei, L., Zhang, Y., Wang, Z., Zhao, L., Zhang, Y., Lu, X. & Cao, L. (2020). Hype spec al in e sion o soil o ganic ma e
con en based on a combined spec al index model. Senso s, 20(10), 1-17.
Wu, Z., Lei, T., Shen, C., Wang, Z., Cao, D. & Hou, T. (2019). ADMET e alua ion in d ug disco e y. 19. Reliable p edic-
ion o human cy och ome P450 inhibi ion using a i icial in elligence app oaches. Jou nal o Chemical In o -
ma ion and Modeling, 59(11), 4587-4601, h ps://doi.o g/10.1021/acs.jcim.9b00801.
Xiang, Y. (2024). Enhancing non-in asi e colo ec al cance sc eening wi h s ool DNA me hyla ion ma ke s and
Ligh GBM machine lea ning, Resea ch Squa e, 1, 1-19, h ps://doi.o g/10.21203/ s.3. s-3857174/ 1.
Xiao, D., Chen, J., Zhang, K. & Qian, H. (2020). P i acy-p ese ing locally weigh ed linea eg ession o e enc yp ed
millions o da a. IEEE Access, 8, 2247-2257, h ps://doi.o g/10.1109/ACCESS.2019.2962700.
Xin, S. & Khalid, K. (2018). Modelling house p ice using idge eg ession and lasso eg ession. In e na ional Jou nal o
Enginee ing & Technology, 7(4), 498-501.
Yao, S., Wu, Q., Kang, Q., Chen, Y. & Yi, L. (2023). An in e p e able XGBoos -based app oach o A c ic na iga ion isk
assessmen . Risk Analysis, 44(2), 459-476, h ps://doi.o g/10.1111/ isa.14175.
Yin, S., Ouyang, P., Xu, D., Liu, L. & Wei, S. (2017). An AdaBoos -based ace de ec ion sys em using pa allel con igu a-
ble a chi ec u e wi h op imized compu a ion. IEEE Sys ems Jou nal, 11(1), 260-271.
Yoo, H., Lee, K., Woo, J., Pa k, S., Lee, S., Joo, J., Bae, J.-S., Hwong, H.-J. & Pa k, B. (2022). A genome-wide associa ion
s udy and machine-lea ning algo i hm analysis on he p edic ion o acial pheno ypes by geno ypes in Ko ean
women. Clinical, Cosme ic and In es iga ional De ma ology, 15, 433-445.
Ze n, A., B oelemann, K. & Kasneci, G. (2023). In e en ional SHAP alues and in e ac ion alues o piecewise linea
eg ession ees. P oceedings o he AAAI Con e ence on A i icial In elligence, 37(9), 11164-11173.
Zhan, C., Zheng, Y., Zhang, H. & Wen, Q. (2021). Random-Fo es -Bagging B oad Lea ning Sys em wi h applica ions o
COVID-19 pandemic. IEEE In e ne o Things Jou nal, 8, 15906-15918.
Zhang, B., Se hy, A., Saina h, T.N. & Ramabhad an, B. (2011). Applica ion speci ic loss minimiza ion using g adien
boos ing. In: IEEE In e na ional Con e ence on Acous ics, Speech and Signal P ocessing (pp. 4880-4883). IEEE,
New Je sey, h ps://doi.o g/10.1109/ICASSP.2011.5947449.
Zhang, J. (2024). Op imiza ion and applica ion o XGBoos logging p edic ion model o po osi y and pe meabili y
based on k-means me hod. Applied Sciences, 14(10), 1-18, h ps://doi.o g/10.3390/app14103956.
Cansu E genç, Ra e Ak aş
Sec o -speci ic inancial o ecas ing wi h machine lea ning algo i hm and SHAP
in e ac ion alues Financial In e ne Qua e ly 2025, ol. 21 / no. 1
Zhang, J. (2024). P edic ion o comp essi e s eng h o geopolyme conc e e landscape design: Applica ion o he
no el hyb id RF-GWO-XGBoos algo i hm. Buildings, 14(3), 1-32, h ps://doi.o g/10.3390/buildings14030591.
Zhang, J., Mucs, D., No inde , U. & S ensson, F. (2019). Ligh GBM: An e ec i e and scalable algo i hm o p edic ion
o chemical oxici y-applica ion o he Tox21 and mu agenici y da a se s. Jou nal o Chemical In o ma ion and
Modeling, 59(10), 4150-4158, h ps://doi.o g/10.1021/acs.jcim.9b00633.
Zhang, P., Jia, Y. & Shang, Y. (2022). Resea ch and applica ion o XGBoos in imbalanced da a. In e na ional Jou nal o
Dis ibu ed Senso Ne wo ks, 18(6), 1-10, h ps://doi.o g/10.1177/15501329221106935.
Zhao, G., Wang, Y. & Wang, J. (2023). In usion de ec ion model o In e ne o Things based on Ligh GBM. IEICE
T ansac ions on Communica ions, 106(8), 622-634, h ps://doi.o g/10.1587/ anscom.2022EBP3169.
Zhou, L., Pan, S., Wang, J. & Vasilakos, A.V. (2017). Machine lea ning on big da a: Oppo uni ies and challenges. Neu-
ocompu ing, 237, 350-361, h ps://doi.o g/10.1016/j.neucom.2017.01.026.
Appendix 1: Pe o mances o machine lea ning models o e es sample
Ac ual s. P edic ed To al Income (Linea Reg ession) Ac ual s. P edic ed To al Income (Ridge Reg ession)
Ac ual s. P edic ed To al Income (Lasso Reg ession) Ac ual s. P edic ed To al Income (Decision T ee)
Ac ual s. P edic ed To al Income (Bagging) Ac ual s. P edic ed To al Income (AdaBoos )
Cansu E genç, Ra e Ak aş
Sec o -speci ic inancial o ecas ing wi h machine lea ning algo i hm and SHAP
in e ac ion alues Financial In e ne Qua e ly 2025, ol. 21 / no. 1
Sou ce: Au ho ’s own wo k.
Ac ual s. P edic ed To al Income (G adien Boos ing) Ac ual s. P edic ed To al Income (XGBoos )
Ac ual s. P edic ed To al Income (Random Fo es )
Cansu E genç, Ra e Ak aş
Sec o -speci ic inancial o ecas ing wi h machine lea ning algo i hm and SHAP
in e ac ion alues Financial In e ne Qua e ly 2025, ol. 21 / no. 1
Appendix 2: SHAP ea u e impo ance and summa y o he inancial o ecas ing esul s o he selec ed machine lea ning models
(a) Linea Reg ession
(b) Ridge Reg ession
Fea u e Value Fea u e Value
High
Low
SHAP Value (impac on he model’s
ou pu )
Mean (|SHAP Values|) (a e age impac (magni ude)
o each ea u e on he model’s ou pu )
SHAP Value (impac on he model’s
ou pu )
Mean (|SHAP Values|) (a e age impac (magni ude)
o each ea u e on he model’s ou pu )
Cansu E genç, Ra e Ak aş
Sec o -speci ic inancial o ecas ing wi h machine lea ning algo i hm and SHAP
in e ac ion alues Financial In e ne Qua e ly 2025, ol. 21 / no. 1
(c) Lasso Reg ession
(d) Decision T ee
Fea u e Value
Fea u e Value
Low
High
Low
High
SHAP Value (impac on he model’s
ou pu )
Mean (|SHAP Values|) (a e age impac (magni ude)
o each ea u e on he model’s ou pu )
SHAP Value (impac on he model’s
ou pu )
Mean (|SHAP Values|) (a e age impac (magni ude)
o each ea u e on he model’s ou pu )
Cansu E genç, Ra e Ak aş
Sec o -speci ic inancial o ecas ing wi h machine lea ning algo i hm and SHAP
in e ac ion alues Financial In e ne Qua e ly 2025, ol. 21 / no. 1
(e) Bagging
(g) AdaBoos
Fea u e Value Fea u e Value
SHAP Value (impac on he model’s
ou pu )
Mean (|SHAP Values|) (a e age impac (magni ude)
o each ea u e on he model’s ou pu )
SHAP Value (impac on he model’s
ou pu )
Mean (|SHAP Values|) (a e age impac (magni ude)
o each ea u e on he model’s ou pu )
Cansu E genç, Ra e Ak aş
Sec o -speci ic inancial o ecas ing wi h machine lea ning algo i hm and SHAP
in e ac ion alues Financial In e ne Qua e ly 2025, ol. 21 / no. 1
( ) Random Fo es
(h) G adien Boos ing
Fea u e Value Fea u e Value
SHAP Value (impac on he model’s
ou pu )
Mean (|SHAP Values|) (a e age impac (magni ude)
o each ea u e on he model’s ou pu )
SHAP Value (impac on he model’s
ou pu )
Mean (|SHAP Values|) (a e age impac (magni ude)
o each ea u e on he model’s ou pu )
Cansu E genç, Ra e Ak aş
Sec o -speci ic inancial o ecas ing wi h machine lea ning algo i hm and SHAP
in e ac ion alues Financial In e ne Qua e ly 2025, ol. 21 / no. 1
Sou ce: Au ho ’s own wo k.
(i) Ligh GBM
(j) XGBoos
Fea u e Value Fea u e Value
SHAP Value (impac on he model’s
ou pu )
Mean (|SHAP Values|) (a e age impac (magni ude)
o each ea u e on he model’s ou pu )
SHAP Value (impac on he model’s
ou pu )
Mean (|SHAP Values|) (a e age impac (magni ude)
o each ea u e on he model’s ou pu )
