Clus e ing o Decision Suppo in he Fashion Indus y
Ana Lisa Amo im do Mon e
O ien ação: P o . D . Ca los Soa es
Co-o ien ação: P o . D . Ped o B i o
Disse ação de Mes ado em Economia
e
Adminis ação de Emp esas
Se emb o, 2012
CLUSTERING FOR DECISION SUPPORT IN THE
FASHION INDUSTRY
po
Ana Lisa Amo im do Mon e
Disse ação de Mes ado em Economia e Adminis ação de Emp esas
O ien ada po
P o . D . Ca los Soa es
Co-o ien ada po
P o . D . Ped o Quelhas B i o
2012
III
No a Biog á ica
Ana Lisa Amo im do Mon e nasceu a 8 de No emb o de 1981 na cidade da
Pó oa de Va zim. Licenciou-se em Ges ão na Faculdade de Economia da Uni e sidade
do Po o (FEP) em 2005. Du an e o cu so pa icipou no 44 h Eu opean Cong ess o
Eu opean Regional Science Associa ion (ERSA) como Assis en e em 2004.
Após e e minado a licencia u a, iniciou em 2006 um es ágio p o issional do
IEFP na emp esa Mon e-SGPS, S.A. na Pó oa de Va zim na á ea de ges ão. Concluído
o es ágio em 2007, ing essou du an e um pe íodo de 2 anos no Banco Popula Po ugal,
S.A., onde execu ou unções de Caixa e de Ges o de Clien es Pa icula es numa das
suas agências em Po ugal. Foi du an e a sua pe manência no banco, que decidiu
in eg a o Mes ado em Economia e Adminis ação de Emp esas na FEP, endo iniciado
no ano lec i o 2009/2010.
Em 2009, inda a sua passagem pela á ea da banca, ing essa numa expe iência
in e nacional. Pa icipa no P og ama de Es ágios In e nacionais da AICEP, o “Ino
Con ac o”, a a és do qual colabo a, du an e um pe íodo de seis meses, com a emp esa
Sonae Sie a, nos seus esc i ó ios localizados em A enas na G écia, na á ea de
Repo ing Ope acional.
Te minado esse es ágio in e nacional em 2010, eg essa a Po ugal e nesse
mesmo ano inicia um es ágio in e no, com du ação de um ano, no Sha ed Se ice
Cen e da emp esa alemã In ineon Tecnologies, si uada no TECMAIA - Pa que de
Ciência e Tecnologia da Maia - onde in eg a o depa amen o de Accoun s Payable.
Em inais de 2011, e minado esse es ágio in e no, a sua dedicação ocou-se na
inalização do Mes ado, endo já concluído a pa e cu icula .

V
Ag adecimen os
Es e abalho só oi possí el ealiza a a és da colabo ação e incen i o de um
conjun o de pessoas e en idades.
Ag adeço ao meu o ien ado , o P o esso Dou o Ca los Manuel Milhei o de
Oli ei a Pin o Soa es, pela a enção, disponibilidade, simpa ia, dinamismo, sen ido
c í ico cons u i o e acima de udo pela con iança que deposi ou nes e p ojec o e na
minha pessoa.
Ao meu co-o ien ado , o P o esso Dou o Ped o Manuel dos San os Quelhas
Tauma u go de B i o, que acei ou con ibui nes e p ojec o com a sua p eciosa ajuda e
disponibilidade e com os seus incon es á eis conhecimen os na á ea que mais domina.
Ao LIAAD - Labo a o y o A i icial In elligence and Decision Suppo , unidade
associada do INESC TEC - INESC Tecnologia e Ciência, que me acolheu
simpa icamen e cedendo o seu espaço e meios pa a pode ealiza es e abalho.
À emp esa Bi olino que disponibilizou os seus dados que se i am de base em
odo es e abalho, pela sua con ibuição e simpa ia.
The esea ch leading o hese esul s has ecei ed unding om he Eu opean
Union Se en h F amewo k P og amme (FP7/2007-2013) unde g an ag eemen n°
260169 (P ojec CoReNe - www.co ene -p ojec .eu).
A odos os meus amigos e colegas, à minha amília e ao meu namo ado que
semp e me apoia am e incen i a am a não desis i e lu a pelos meus objec i os, que
semp e me ize am ac edi a nas minhas capacidades.
VII
Resumo
O ema des e abalho é a segmen ação das encomendas de uma emp esa belga, a
Bi olino, cuja ac i idade se oca essencialmen e na come cialização de camisas ei as à
medida de cada clien e. Essa segmen ação é ei a com a ajuda de écnicas de Da a
Mining, endo sido u ilizada nes e abalho a écnica de Clus e ing, a qual é conside ada
uma das mais impo an es écnicas de Da a Mining. Es a écnica az a pa ição dos
dados de aco do com um dado c i é io de simila idade ou dis ância, sendo a Dis ância
Euclidiana o mais comummen e u ilizado.
O mé odo de Clus e ing seleccionado pa a aze a pa ição dos dados oi o K-
Medoids, dado es e se menos sensí el a ou lie s do que ou os mé odos e
p incipalmen e po pode lida com dados nominais. A compu ação des e mé odo oi
ealizada com ecu so a um so wa e pa a Da a Mining, o chamado RapidMine . Das
á ias expe iências e ec uadas com os dados no RapidMine , o am seleccionados os
esul ados que se iam objec o de análise e in e p e ação que numa pe spec i a écnica
que numa pe spec i a de Ma ke ing.
Os esul ados mos am que é possí el iden i ica as endências de moda nas
camisas da Bi olino pa a apoia as decisões da emp esa a ní el do Design e do
Ma ke ing. Es e abalho con ibui pa a demons a a po encialidade da u ilização de
e amen as de Da a Mining pa a analisa g andes quan idades de dados de emp esas
ans o mando-os em in o mação ú il e daí ex aí em conhecimen o ace ca do seu
negócio. Se á esse conhecimen o que lhes ai pe mi i oma impo an es decisões em
empo ú il e ob e an agens compe i i as.
Pala as-cha e: Da a Mining, Clus e ing, K-Means, K-Medoids, Ma ke ing,
Segmen ação

XV
Lis o Tables
Table 2.1 – Da a Mining Tasks.........................................................................................8
Table 3.1 – Clus e ing Pe spec i es................................................................................13
Table 3.2 – K-Means: Common choices o p oximi y, cen oids and objec i e
unc ions..........................................................................................................................19
Table 3.3 – Selec ed Dis ance Func ions be ween Pa e ns x and y...............................20
Table 3.4 – In e nal Validi y Indices..............................................................................22
Table 3.5 – Ex e nal Validi y Indices.............................................................................24
Table 4.1 – Requi emen s o an E ec i e Segmen a ion..............................................28
Table 4.2 – S eps in Segmen a ion P ocess....................................................................29
Table 4.3 – Le els o Ma ke ing Segmen a ion..............................................................30
Table 4.4 – Classi ica ion o Segmen a ion Bases..........................................................32
Table 4.5 – E alua ion o Segmen a ion Bases..............................................................34
Table 4.6 – Classi ica ion o Segmen a ion Me hods.....................................................35
Table 4.7 – E alua ion o Segmen a ion Me hods..........................................................37
Table 5.1 –Shi s A ibu es Values o Clus e 1............................................................47
Table 5.2 – Rep esen a ion o Binominal A ibu es......................................................54
Table 5.3 – Shi s A ibu es Values o Clus e 2...........................................................55
Table 5.4 – Simila i ies agains Clus e 1.......................................................................56
Table 5.5 – Shi s A ibu es Values o Clus e 3...........................................................57
Table 5.6 – Simila i ies agains Clus e 1 and Clus e 2.................................................58
Table 5.7 – Shi s A ibu es Values o Clus e 4...........................................................59
Table 5.8 – Simila i ies agains Clus e 1, Clus e 2 and Clus e 3................................60
Table 5.9 – Shi s A ibu es Values o Clus e 5...........................................................61
Table 5.10 – Simila i ies agains Clus e 1, Clus e 2, Clus e 3 and Clus e 4..............62
Table 5.11 – Shi s A ibu es Values o Clus e 6.........................................................63
Table 5.12 – Simila i ies agains Clus e 1, Clus e 2, Clus e 3, Clus e 4 and
Clus e 5...........................................................................................................................64
Table 6.1 – Rep esen a ion o Dis inc A ibu es Values o Segmen 2........................69
Table 6.2 –Rep esen a ion o Bi olino O de s by Gende .............................................69
Table 6.3 – Classi ica ion o Segmen a ion Va iables....................................................72
Table 6.4 – Cus ome s A ibu es Values o Segmen 1.................................................73
XVI
Table 6.5 – Obesi y o Male Cus ome s measu ed by he Colla Size...........................77
Table 6.6 – Obesi y o Male Cus ome s.........................................................................78
Table 6.7 – Bi olino Gi Vouche s Usage analysis.......................................................79
Table 6.8 – Cus ome s A ibu es Values o Segmen 2.................................................80
Table 6.9 – Obesi y o Female Cus ome s measu ed by he Colla Size........................84
Table 6.10 - Obesi y o Female Cus ome s....................................................................84
Table 6.11 - Cus ome s A ibu es Values o Segmen 3................................................85
Table 6.12 - Simila i ies agains Segmen 1...................................................................85
Table 6.13 - Cus ome s A ibu es Values o Segmen 4................................................86
Table 6.14 - Simila i ies agains Segmen 1 and Segmen 3...........................................87
Table 6.15 - Cus ome s A ibu es Values o Segmen 5................................................88
Table 6.16 - Simila i ies agains Segmen 1, Segmen 3 and Segmen 4........................88
Table 6.17 - Cus ome s A ibu es Values o Segmen 6................................................90
Table 6.18 - Simila i ies agains Segmen 1, Segmen 3, Segmen 4 and Segmen 5.....90
XVII
Lis o Figu es
Figu e 2.1 – The ou s ages o Da a Mining P ocess......................................................9
Figu e 3.1 – Hie a chical Clus e ing..............................................................................16
Figu e 3.2 – Pa i ional Clus e ing..................................................................................16
Figu e 3.3 – A Taxonomy o Clus e ing App oaches.....................................................16
Figu e 3.4 – K-Means Clus e ing S eps..........................................................................18
Figu e 3.5 – K-Means algo i hm P ocess.......................................................................19
Figu e 3.6 – A Simpli ied Classi ica ion o Valida ion Techniques..............................26
Figu e 4.1 – Le els o Ma ke ing Segmen a ion............................................................30
Figu e 4.2 – Classi ica ion o Clus e ing Me hods.........................................................39
Figu e 4.3 – Clus e ing Me hods: (a) nono e lapping, (b) o e lapping, (c) uzzy........39
Figu e 4.4 – Nono e lapping Hie a chical Clus e ing Me hods....................................40
Figu e 5.1 – Resul o K-Medoids Clus e ing Ex ac Clus e P o o ype o k=6..........45
Figu e 5.2 – Rep esen a ion o Clus e 1........................................................................46
Figu e 5.3 – Illus a ion o Designing a Men Shi on Bi olino websi e – Shi
a ibu es..........................................................................................................................48
Figu e 5.4 – Illus a ion o Designing a Men Shi on Bi olino websi e – Cus ome s
a ibu es..........................................................................................................................49
Figu e 5.5 – Fab ics o Men Shi s.................................................................................49
Figu e 5.6 – Fab ic Colo s o Men Shi s.......................................................................50
Figu e 5.7 – Fab ic Design and Pa e n o Men Shi s...................................................50
Figu e 5.8 – Fab ic Finish o Men Shi s........................................................................51
Figu e 5.9 – Fab ic S uc u e o Men Shi s...................................................................51
Figu e 5.10 – Fab ic Ma e ial o Men Shi s..................................................................52
Figu e 5.11 – Colla o Men Shi s.................................................................................52
Figu e 5.12 – Cu o Men Shi s....................................................................................53
Figu e 5.13 – Placke o Men Shi s...............................................................................53
Figu e 5.14 – Hem o Men Shi s...................................................................................54
Figu e 5.15 – Rep esen a ion o Clus e 2......................................................................55
Figu e 5.16 – Rep esen a ion o Clus e 3......................................................................57
Figu e 5.17 – Rep esen a ion o Clus e 4......................................................................59
Figu e 5.18 – Rep esen a ion o Clus e 5......................................................................61
XVIII
Figu e 5.19 – Rep esen a ion o Clus e 6......................................................................63
Figu e 6.1 – Resul o K-Medoids Clus e ing Ex ac Clus e P o o ype o k=6 wi h
ex a a iables..................................................................................................................67
Figu e 6.2 – Rep esen a ion o Segmen 1......................................................................73
Figu e 6.3 – Rep esen a ion o Men O de s by coun y.................................................74
Figu e 6.4 – Age G oups o Male Cus ome s.................................................................74
Figu e 6.5 – Heigh G oups (in cm) o Male Cus ome s................................................75
Figu e 6.6 – Weigh G oups (in kg) o Male Cus ome s................................................75
Figu e 6.7 – BMI measu es o Male Cus ome s.............................................................76
Figu e 6.8 – Colla G oups o Male Cus ome s.............................................................77
Figu e 6.9 – Rep esen a ion o Segmen 2......................................................................78
Figu e 6.10 – Bi olino Gi Vouche s............................................................................80
Figu e 6.11 – Rep esen a ion o Women O de s by Coun y.........................................81
Figu e 6.12 – Age G oups o Female Cus ome s...........................................................81
Figu e 6.13 – Heigh G oups (in cm) o Female Cus ome s..........................................82
Figu e 6.14 – Weigh G oups (in kg) o Female Cus ome s..........................................82
Figu e 6.15 – BMI measu es o Female Cus ome s.......................................................83
Figu e 6.16 – Colla G oups o Female Cus ome s........................................................83
Figu e 6.17 – Rep esen a ion o Segmen 3....................................................................84
Figu e 6.18 – Rep esen a ion o Segmen 4....................................................................86
Figu e 6.19 - Rep esen a ion o Segmen 5....................................................................87
Figu e 6.20 - Rela ion be ween Age and BMI o Male Cus ome s...............................89
Figu e 6.21 - Rep esen a ion o Segmen 6....................................................................89
Figu e 6.22 - Rela ion be ween Coun y and A ilia e o Male cus ome s...................91
Figu e 6.23 - Rela ion be ween Fi and Age o Male Cus ome s..................................92
Figu e 6.24 - Fi choices o Men Shi s..........................................................................92
Figu e 6.25 - Rela ion be ween Pocke and Age o Male Cus ome s...........................93
Figu e 6.26 - Type o Pocke o Men Cus ome s..........................................................93
Figu e 6.27 - To al O de s by Coun y o Bi olino Shi s in 2011................................94
Figu e 6.28 - To al O de s by A ilia e o Bi olino Shi s in 2011................................94
XIX
Abb e ia ions
AID – Au oma ic In e ac ion De ec ion
ANN – A i icial Neu al Ne wo k
BMI – Body Mass Index
CART – Classi ica ion and Reg ession T ees
CHAID – AID o Ca ego ical Dependen Va iables
CRISP-DM – C oss Indus y S anda d P ocess o Da a Mining
DM – Da a Mining
KD – Knowledge Disco e y
MAID - AID o Mul iple Dependen Va iables

PART I
“Theo y helps us o bea ou igno ance o ac s.”
By Geo ge San ayana, a Spanish philosophe
3
1. INTRODUCTION
“The e a e no sec e s o success.
I is he esul o p epa a ion, ha d wo k,
and lea ning om ailu e.”
By Colin Powell, an Ame ican s a esman
This disse a ion is ocused on he p oblem o suppo ing ashion indus y in i s
p oduc ion/design and ma ke ing decisions based on Da a Mining (DM) app oaches.
Sho li e cycles, high ola ili y, low p edic abili y, and high impulse pu chasing is
being appoin ed has cha ac e is ics o ashion indus y (Lo e al, 2008).
The da a ha companies collec abou hei cus ome s is one o i s g ea es asse s
(Ahmed, 2004). Howe e , companies inc easingly end o accumula e huge amoun s o
cus ome da a in la ge da abases (Shaw e al, 2001) and wi hin his as amoun o da a
is all so s o aluable in o ma ion ha could make a signi ican di e ence o he way in
which any company un hei business, and in e ac wi h hei cu en and p ospec i e
cus ome s and gaining compe i i e edge on hei compe i o s (Ahmed, 2004).
Gi en ha and he ac ha companies ha e o be able o eac apidly o he
changing ma ke demands bo h locally and globally (Ahmed, 2004), i is u gen ha
hey manage e icien ly he in o ma ion abou hei cus ome s. So, companies can
u ilize DM echniques o ex ac he unknown and po en ially use ul in o ma ion abou
cus ome cha ac e is ics and hei pu chase pa e ns (Shaw e al, 2001). DM ools can,
hen, p edic u u e ends and beha io s, allowing businesses o make knowledge-
d i en decisions ha will a ec he company, bo h sho e m and long e m (Ahmed,
2004).
DM is also being used in e-comme ce indus y o s udy and iden i y he
pe o mance limi a ions and o analyze da a o pa e ns and, a he same ime, is
helping i o inc ease sale and emo e poli ical and physical bounda ies (Ahmed, 2004).
The iden i ica ion o such pa e ns in da a is he i s s ep o gaining use ul ma ke ing
insigh s and making c i ical ma ke ing decisions (Shaw e al, 2001). In oday’s
en i onmen o complex and e e changing cus ome p e e ences, ma ke ing decisions
10
2.3.2. T ans o m Da a using Da a Mining Techniques
DM akes da a, business oppo uni ies and p oduces ac ionable esul s o he
Take Ac ion s age. Howe e , i is impo an o unde s and wha esul s i mus gi e o
make he DM p ocess success ul and also o pay a en ion o he nume ous p oblems
ha can in e e e wi h i . Some usual p oblems a e, o example, bad da a o ma s,
con using da a ields, lack o unc ionali y, imeliness (Be y and Lino , 1997).
2.3.3. Take Ac ion
This is whe e he esul s om DM a e ac ed upon and esul s a e ed in o he
Measu emen s age. The ques ion he e is how o inco po a e in o ma ion in o business
p ocesses. The business p ocesses mus p o ide he da a eedback needed o he DM
p ocess be success ul (Be y and Lino , 1997).
2.3.4. Measu e he Resul s
Measu emen p o ides he eedback o con inuously imp o ing esul s.
Measu emen he e e e s speci ically o measu es o business alue ha go beyond
esponse a es and cos s, beyond a e ages and s anda d de ia ions. The ques ion he e is
wha o measu e and how o app oach he measu emen so i p o ides he bes inpu o
u u e use. The speci ic measu emen s needed depend on a numbe o ac o s: he
business oppo uni y, he sophis ica ion o he o ganiza ion, pas his o y o
measu emen s ( o end analysis), and he a ailabili y o da a. We can see ha his
s age depends c i ically on in o ma ion p o ided in he p e ious s ages (Be y and
Lino , 1997).
In sho , DM p ocess places DM in o a con ex o c ea ing business alue. In
he business wo ld, DM p o ides he abili y o op imize decision-making using
au oma ed me hods o lea n om pas ac ions. Di e en o ganiza ions adap DM o hei
own en i onmen , in hei own way (Be y and Lino , 1997).

11
2.4. Da a Mining Me hodology
DM me hodology e e s o s age wo o DM p ocess, i.e., he s age whe e he
ac ual mining o da a akes place and whe e he esul ing in o ma ion is smel ed o
p oduce knowledge (Be y and Lino , 1997).
The e a e, essen ially, wo basic app oaches o DM, he hypo hesis es ing and
he knowledge disco e y. The i s is a op-down app oach ha a emp s o subs an ia e
o disp o e p econcei ed ideas. The second is a bo om-up app oach ha s a s wi h he
da a and ies o ge i o ell us some hing we didn´ al eady know (Be y and Lino ,
1997).
2.4.1. Hypo hesis Tes ing
A hypo hesis is a p oposed explana ion whose alidi y can be es ed. Tes ing he
alidi y o a hypo hesis is done by analyzing da a ha may simply be collec ed by
obse a ion o gene a ed h ough an expe imen , such as es mailing (Be y and Lino ,
1997).
The p ocess o hypo heses es ing comp ises he ollowing s eps: (1) Gene a e
good ideas (hypo hesis); (2) De e mine wha da a would allow hese hypo heses o be
es ed; (3) Loca e da a; (4) P epa e da a o analysis; (5) Build compu e models based
on he da a; (6) E alua e compu e models o con i m o ejec hypo heses (Be y and
Lino , 1997).
2.4.2. Knowledge Disco e y
Knowledge disco e y can be ei he di ec ed o undi ec ed. In he i s , he ask is
o explain he alue o some pa icula ield in e ms o all he o he s; we selec he
a ge ield and di ec he compu e o ell us how o es ima e, classi y, o p edic i . In
he second, he e is no a ge ield; we ask he compu e o iden i y pa e ns in he da a
ha may be signi ican (Be y and Lino , 1997). In o he wo ds, we use undi ec ed KD
o ecognize ela ionships in he da a and di ec ed KD o explain hose ela ionships
once hey ha e been ound (Be y and Lino , 1997).
12
2.4.2.1. Di ec ed Knowledge Disco e y
Di ec ed KD is cha ac e ized by he p esence o a single a ge ield whose alue
is o be p edic ed in e ms o he o he ields in he da abase (Be y and Lino , 1997). I
is he p ocess o inding meaning ul pa e ns in da a ha explain pas e en s in such a
way we can use he pa e ns o help p edic u u e e en s (Be y and Lino , 1997).
The s eps o his p ocess a e: (1) Iden i y sou ces o p eclassi ied da a; (2)
P epa e da a o analysis; (3) Build and ain a compu e model; and (4) E alua e he
compu e model (Be y and Lino , 1997).
2.4.2.2. Undi ec ed Knowledge Disco e y
Undi ec ed KD is di e en om di ec ed KD, in he way ha he e is no a ge
ield. The DM ool is simply le loose on he da a in he hope ha i will disco e
meaning ul s uc u e (Be y and Lino , 1997).
The p ocess o undi ec ed KD ha e he ollowing s eps: (1) Iden i y sou ces o
da a; (2) P epa e da a o analysis; (3) Build and ain a compu e model; (4) E alua e
he compu e model; (5) Apply compu e model o new da a; (6) Iden i y po en ial
a ge s o di ec ed KD; and (7) Gene a e new hypo hesis o es (Be y and Lino ,
1997).
In his wo k he DM me hodology used was he undi ec ed KD, mo e p ecisely,
he CRISP-DM me hodology. This me hodology and i s di e en s eps a e desc ibed in
Annex B.
13
3. CLUSTERING
“Science ne e sol es a p oblem
wi hou c ea ing en mo e.”
By Geo ge Be na d Shaw, an I ish d ama is
3.1. Clus e ing De ini ion
Acco ding o Velmu ugan and San hanam (2010) clus e ing consis s o “c ea ing
g oups o objec s based on hei ea u es in such a way ha he objec s belonging o he
same g oups a e simila and hose belonging o di e en g oups a e dissimila .”
In ope a ional e ms, clus e ing can be de ined as ollows: “Gi en a
ep esen a ion o n objec s, ind k g oups based on a measu e o simila i y such ha he
simila i ies be ween objec s in he same g oup a e high while he simila i ies be ween
objec s in di e en g oups a e low” (Jain, 2009).
Fo Mi kin (2005), clus e ing is “a discipline on he in e sec ion o di e en
ields and can be iewed om di e en angles” and he e o e he inds i use ul o
dis inguish he di e en pe spec i es o s a is ics, machine lea ning, da a mining and
classi ica ion abou clus e ing, which a e p esen ed on Table 3.1.
Pe spec i e
Desc ip ion
S a is ics
Tends o iew any da a able as a sample om a p obabili y
dis ibu ion whose p ope ies o pa ame e s a e o be es ima ed
wi h he da a.
Machine lea ning
Tends o iew he da a as a de ice o lea ning how o p edic
p e-speci ied o newly c ea ed ca ego ies.
Da a mining
Assumes ha a da ase o a da abase has al eady been
collec ed and he majo conce n is in inding pa e ns and
egula i ies wi hin he da a as hey a e, despi e how bad o
good i e lec s he p ope ies o he phenomenon in ques ion.
Classi ica ion/
Is an ac ual o ideal a angemen o en i ies unde
14
Pe spec i e
Desc ip ion
Knowledge-disco e y
conside a ion in classes o: (1) shape and keep knowledge; (2)
cap u e he s uc u e o phenomena; and (3) ela e di e en
aspec s o a phenomenon in ques ion o each o he .
Table 3.1 – Clus e ing Pe spec i es
3.2. Clus e ing Goals
Clus e ing goals a e ypes o p oblems o da a analysis o which clus e ing can
be applied. Mi kin (2005) enume a es some objec i es o clus e ing, which a e no
mu ually exclusi e no do hey co e he en i e ange o clus e ing goals. They a e:
1. S uc u ing, ha is ep esen ing da a as a se o g oups o simila objec s.
2. Desc ip ion o clus e s in e ms o ea u es, no necessa ily in ol ed in inding he
clus e s.
3. Associa ion, which is inding in e ela ions be ween di e en aspec s o a
phenomenon by ma ching clus e desc ip ions in spaces co esponding o he
aspec s.
4. Gene aliza ion, ha is making gene al s a emen s abou da a and, po en ially, he
phenomena he da a ela e o.
5. Visualiza ion, which is ep esen ing clus e s uc u es as isual images.
3.3. Clus e ing S ages
The clus e ing p ocess is simila o DM p ocess desc ibed on sec ion 2.3,
howe e we ind i use ul o dis inguish he s ages o he clus e ing p ocess. Acco ding
o Mi kin (2005), clus e ing, as a DM ac i i y, ypically in ol es he ollowing s ages:
1. De eloping a da ase
2. Da a p e-p ocessing and s anda dizing
3. Finding clus e s in da a
4. In e p e a ion o clus e s
5. D awing conclusions
In he i s s age i is necessa y o de elop a subs an i e p oblem o issue and
hen de e mine wha da ase , ela ed o he issue, can be collec ed om an exis ing
da abase o se o expe imen s o su eys, e c.
15
The second is he s age o p epa ing da a p ocessing by a clus e ing algo i hm. I
no mally includes de eloping a uni o m da ase om a da abase, checking o missing
o un eliable en ies, escaling and s anda dizing a iables, de i ing a uni ied simila i y
measu e, e c.
Finding clus e s in da a is he hi d s age and in ol es he applica ion o a
clus e ing algo i hm which esul s in a clus e s uc u e
1
o be p esen ed, along wi h
in e p e a ion aids, o subs an i e specialis s o an expe judgmen and in e p e a ion.
A his s age, he expe may no see ele ance in he esul s and sugges a modi ica ion
o he da a by adding/ emo ing ea u es and/o en i ies. The modi ied da a is subjec o
he same p ocessing p ocedu e.
The inal s age is d awing conclusions om he in e p e a ion o he esul s
ega ding he issue in ques ion. The mo e ocused he egula i ies a e implied in he
indings, he be e he quali y o conclusions a e.
3.4. Clus e ing Algo i hms
Clus e ing algo i hms can be di ided in wo g oups: (1) hie a chical and (2)
pa i ional. Hie a chical clus e ing algo i hms (Figu e 3.1) ecu si ely ind nes ed
clus e s ei he in agglome a i e mode (s a ing wi h each da a poin in i s own clus e
and me ging he mos simila pai o clus e s successi ely o o m a clus e hie a chy)
o in di isi e ( op-down) mode (s a ing wi h all he da a poin s in one clus e and
ecu si ely di iding each clus e in o smalle clus e s). Pa i ional clus e ing algo i hms
(Figu e 3.2) ind all he clus e s simul aneously as a pa i ion o he da a and do no
impose a hie a chical s uc u e (Jain, 2009).
Inpu o a hie a chical algo i hm is a n x n simila i y ma ix, whe e n is he
numbe o objec s o be clus e ed while a pa i ional algo i hm can ei he use a n x d
pa e n ma ix, whe e n objec s a e embedded in a d-dimensional ea u e space, o a n x
n simila i y ma ix (Jain, 2009)
The mos well-known hie a chical algo i hms a e single-link and comple e-link
(Jain 2009) and he mos popula and simples pa i ional algo i hm is K-Means. Figu e
3.3 syn he izes hese clus e ing app oaches.
1
Acco ding o Mi kin (2005), a clus e s uc u e is “a ep esen a ion o an en i y se I as a se o clus e s

16
Figu e 3.1 – Hie a chical Clus e ing
Figu e 3.2 – Pa i ional Clus e ing
Figu e 3.3 – A Taxonomy o Clus e ing App oaches (Jain e al, 2000)
17
3.4.1. K-Means Clus e ing
K-Means is one o he mos widely used algo i hms o clus e ing. The main
easons o i s popula i y a e: (1) ease o implemen a ion, (2) simplici y, (3) e iciency,
and (4) empi ical success (Jain, 2009). Likewise, Mi kin (2005) inds K-Means
compu a ionally easy, as and memo y-e icien . Howe e , he poin s ou some
p oblems ela ed o he ini ial se ing and s abili y o esul s.
Mi kin (2005) de ines and esumes K-Means as “a majo clus e ing me hod
p oducing a pa i ion o he en i y se in o non-o e lapping clus e s along wi h wi hin-
clus e cen oids
2
. I p oceeds in i e a ions consis ing o wo s eps each: one s ep
upda es clus e s acco ding o he minimum dis ance ule
3
; he o he s ep upda es
cen oids as he cen e s o g a i y o clus e s. The me hod implemen s he so-called
al e na ing minimiza ion algo i hm o he squa e e o c i e ion
4
. To ini ialize he
compu a ions, ei he a pa i ion o a se o all k en a i e cen oids mus be speci ied.”
An example o a K-Means algo i hm is gi en by Jain (2009) and is desc ibed as
ollows:
Le { }, i = 1,…, n be he se o n d-dimensional poin s o be clus e ed in o
a se o K clus e s, { }, k = 1,…, K. K-Means algo i hm inds a pa i ion such ha
he squa ed e o be ween he empi ical mean o a clus e and he poin s in he clus e is
minimized. Le µk be he mean o clus e ck. The squa ed e o be ween µk and he poin s
in clus e ck is de ined as ( ) ∑
. The goal o K-Means is o
minimize he sum o he squa ed e o o e all he k clus e s,
( ) ∑ ∑
. K-Means s a s wi h an ini ial pa i ion wi h K
clus e s and assign pa e ns o clus e s so as o educe he squa ed e o . Since he
squa ed e o always dec eases wi h an inc ease in he numbe o clus e s K (wi h J(C) =
0 when K = n), i can be minimized only o a ixed numbe o clus e s.
2
Acco ding o Mi kin (2005), a cen oid is a mul idimensional ec o minimizing he summa y dis ance
o clus e ’s elemen s. I he dis ance is Euclidean squa ed, he cen oid is equal o he cen e o g a i y o
he clus e .
3
Acco ding o Mi kin (2005), minimum dis ance ule is he ule which assigns each o he en i ies o i s
nea es cen oid.
4
Acco ding o Mi kin (2005), he squa e e o c i e ion is he sum o summa y dis ances om clus e
cen oids, which is minimized by K-means. The dis ance used is he Euclidean dis ance squa ed which is
exp essed by he equa ion ( ) ∑(
) (Has ie e al, 2001).
18
So, he main s eps o K-Means algo i hm a e:
1. Selec an ini ial pa i ion wi h k clus e s ( epea s eps 2 and 3 un il clus e
membe ship s abilizes).
2. Gene a e a new pa i ion by assigning each pa e n o i s closes clus e cen e .
3. Compu e new clus e cen e s.
Figu e 3.4 illus a es his p ocess, as well Figu e 3.5. S a ing wi h wo andom
poin s as cen oids (1), he s age (2) assigns each poin o he clus e nea es o i . In he
s age (3), he associa ed poin s a e a e aged ou o p oduce he new loca ion o he
cen oid, lea ing us wi h he inal con igu a ion (4). A e each i e a ion he inal
con igu a ion is ed back in o he same loop ill he cen oids con e ge.
(1) (2)
(3) (4)
Figu e 3.4 –K-Means Clus e ing S eps
19
Figu e 3.5 – K-Means Algo i hm P ocess
Table 3.2 p esen s some dis ance unc ions, including he Euclidean dis ance
while Table 3.3 indica es he mos common choices o p oximi y, cen oids and
objec i e unc ions specially o K-Means. The Eucliden dis ance is he dis ance me ic
ha we a e going o use in his s udy.
Table 3.2 – Selec ed Dis ance Func ions be ween Pa e ns x and y (Ped ycz, 2005)
26
Name
Exp ession
Desc ip ion
en opy inc eases.
- To compu e he en opy o a
da ase , we need o calcula e he class
dis ibu ion o objec s in each clus e
, whe e he sum is
aken o e all he classes.
- The o al en opy o a se o
clus e s is calcula ed as he weigh ed
sum o he en opies o all clus e s.
- = he size o clus e j
- m = he numbe o clus e s
- n = he o al numbe o da a poin s
Table 3.5 – Ex e nal Validi y Indices (Adap ed om Rendón e al, 2011)
Figu e 3.6 – A Simpli ied Classi ica ion o Valida ion Techniques
(B un e al, 2006)

27
4. SEGMENTATION IN MARKETING
“Do no buy ma ke sha e.
Figu e ou how o ea n i .”
Philip Ko le in Ma ke ing Managemen , 11 h Edi ion
4.1. Segmen a ion De ini ion
Ko le (2005) de ines ma ke segmen a ion as “ he ac o di iding a ma ke in o
smalle g oups o buye s wi h dis inc needs, cha ac e is ics, o beha io s who migh
equi e sepa a e p oduc s and/o ma ke ing mixes.” Despi e ha de ini ion, he e is
ano he one ha is conside ed he bes by ma ke e s gi en by Smi h (1956): “ma ke
segmen a ion in ol es iewing a he e ogeneous ma ke as a numbe o smalle
homogeneous ma ke s, in esponse o di e ing p e e ences, a ibu able o he desi es o
consume s o mo e p ecise sa is ac ion o hei a ying wan s.”
Acco ding o Wedel and Kamaku a (2002), in ma ke segmen a ion one
dis inguishes homogeneous g oups o cus ome s who can be a ge ed in he same
manne because hey ha e simila needs and p e e ences. Fo hem, ma ke segmen s a e
no eal en i ies na u ally occu ing in he ma ke place, bu g oupings c ea ed by
manage s o help hem de elop s a egies ha be e mee consume needs a he highes
expec ed p o i o he company. The e o e, segmen a ion is a e y use ul concep o
manage s.
4.2. Segmen a ion E ec i eness
Acco ding o Ko le (2000), an e ec i e segmen a ion mus mee some c i e ia.
The segmen s mus be measu able ( he size, pu chasing powe , p o iles o segmen s can
be measu e), subs an ial (segmen s mus be la ge o p o i able enough o se e),
accessible (segmen s mus be e ec i ely eached and se ed), di e en iable (segmen s
mus espond di e en ly o di e en ma ke ing mix elemen s and ac ions), and
ac ionable (mus be able o a ac and se e he segmen s).
28
Fo Wedel and Kamaku a (2000) he c i e ia ha de e mine he e ec i eness and
p o i abili y o ma ke ing s a egies a e six and a e desc ibed on Table 4.1.
C i e ia
Desc ip ion
1
Iden i iabili y
Is he ex en o which manage s can ecognize dis inc g oups o
cus ome s in he ma ke place by using speci ic segmen a ion
bases. They should be able o iden i y he cus ome s in each
segmen on he basis o a iables ha a e easily measu ed.
2
Subs an iali y
This c i e ion is sa is ied i he a ge ed segmen s ep esen a la ge
enough po ion o he ma ke o ensu e he p o i abili y o a ge ed
ma ke ing p og ams. In he limi , his c i e ion may be applied o
each indi idual cus ome .
3
Accessibili y
Is he deg ee o which manage s a e able o each he a ge ed
segmen s h ough p omo ional o dis ibu ional e o s.
4
Responsi eness
This c i e ion is sa is ied i he segmen s espond uniquely o
ma ke ing e o s a ge ed o hem. I is c i ical because
di e en ia ed ma ke ing mixes will be e ec i e only i each
segmen is homogeneous and unique in i s esponse o hem.
5
S abili y
I is necessa y, a leas o a pe iod long enough o iden i ica ion
o he segmen s, implemen a ion o he segmen ed ma ke ing
s a egy, and he s a egy o p oduce esul s. Only segmen s ha
a e s able in ime can p o ide he unde lying basis o he
de elopmen o a success ul ma ke ing s a egy.
6
Ac ionabili y
This c i e ion is sa is ied i he iden i ica ion o he segmen s
p o ides guidance o decisions on he e ec i e speci ica ion o
ma ke ing ins umen s. The ocus is on whe he he cus ome s in
he segmen and he ma ke ing mix necessa y o sa is y hei
needs a e consis en wi h he goals and co e compe ences o he
company.
Table 4.1 – Requi emen s o an E ec i e Segmen a ion
29
4.3. Segmen a ion P ocess
Ma ke e s a e inc easingly combining se e al a iables in an e o o iden i y
smalle , be e -de ined a ge g oups. This has led some ma ke esea che s o ad oca e
a needs-based ma ke segmen a ion app oach (Ko le , 2008). Thus a se en-s ep
segmen a ion app oach was p oposed by Bes (2005), which s eps a e desc ibed on
Table 4.2.
S eps
Desc ip ion
1
Needs-based
Segmen a ion
G oup cus ome s in o segmen s based on simila needs and
bene i s sough by cus ome in sol ing a pa icula consump ion
p oblem.
2
Segmen
Iden i ica ion
Fo each needs-based segmen , de e mine which demog aphics,
li es yles, and usage beha io s make he segmen dis inc and
iden i iable (ac ionable).
3
Segmen
A ac i eness
Using p ede e mined segmen a ac i eness c i e ia (such as
ma ke g ow h, compe i i e in ensi y, and ma ke access)
de e mine he o e all a ac i eness o each segmen .
4
Segmen
P o i abili y
De e mine segmen p o i abili y.
5
Segmen
Posi ioning
Fo each segmen , c ea e a “ alue p oposi ion” and p oduc -p ice
posi ioning s a egy based on ha segmen ’s unique cus ome
needs and cha ac e is ics.
6
Segmen
“Acid Tes ”
C ea e “segmen s o yboa d” o es he a ac i eness o each
segmen ’s posi ioning s a egy.
7
Ma ke ing-Mix
S a egy
Expand segmen posi ioning s a egy o include all aspec s o he
ma ke ing-mix: p oduc , p ice, p omo ion and place.
Table 4.2 – S eps in Segmen a ion P ocess
(Adap ed om Ko le , 2008)
30
4.4. Le els o Ma ke Segmen a ion
Because buye s ha e unique needs and wan s, each buye is po en ially a
sepa a e ma ke . Ideally, hen, a selle migh design a sepa a e ma ke ing p og am o
each buye . Howe e , al hough some companies a emp o se e buye s indi idually,
many o he s ace la ge numbe s o smalle buye s and do no ind comple e
segmen a ion wo hwhile. Ins ead, hey look o b oade classes o buye s who di e in
hei p oduc needs o buying esponses. Thus, ma ke segmen a ion can be ca ied ou
a se e al di e en le els (Ko le , 2008).
Figu e 4.1 – Le els o Ma ke ing Segmen a ion
Figu e 4.1 shows ha companies can p ac ice no segmen a ion (mass
ma ke ing), comple e segmen a ion (mic oma ke ing), o some hing in be ween
(segmen ma ke ing o niche ma ke ing).
Le els o Ma ke ing Segmen a ion
Mass Ma ke ing
Same p oduc o all consume s
No
Segmen a ion
Segmen Ma ke ing
Di e en p oduc s o one o mo e
segmen s
Some
Segmen a ion
Niche Ma ke ing
Di e en p oduc s o subg oups wi hin
segmen s
Mo e
Segmen a ion
Mic oma ke ing
P oduc s o sui he as es o indi iduals o
loca ions
Comple e
Segmen a ion
Table 4.3 – Le els o Ma ke ing Segmen a ion
Mass
Ma ke ing Segmen
Ma ke ing Niche
Ma ke ing Mic o
Ma ke ing
31
4.4.1. Mass Ma ke ing
Companies ha e no always p ac iced a ge ma ke ing. In ac , o mos o he
1900s, majo consume p oduc s companies held as o mass ma ke ing—mass
p oducing, mass dis ibu ing, and mass p omo ing abou he same p oduc in abou he
same way o all consume s. The adi ional a gumen o mass ma ke ing is ha i
c ea es he la ges po en ial ma ke , which leads o he lowes cos s, which in u n can
ansla e in o ei he lowe p ices o highe ma gins. Howe e , many ac o s now make
mass ma ke ing mo e di icul . The p oli e a ion o dis ibu ion channels and
ad e ising media has also made i di icul o p ac ice "one-size- i s-all" ma ke ing
(Ko le , 2008).
4.4.2. Segmen Ma ke ing
A company ha p ac ices segmen ma ke ing isola es b oad segmen s ha make
up a ma ke and adap s i s o e s o mo e closely ma ch he needs o one o mo e
segmen s. Segmen ma ke ing o e s se e al bene i s o e mass ma ke ing. The
company can ma ke mo e e icien ly, a ge ing i s p oduc s o se ices, channels, and
communica ions p og ams owa d only consume s ha i can se e bes and mos
p o i ably. The company can also ma ke mo e e ec i ely by ine- uning i s p oduc s,
p ices, and p og ams o he needs o ca e ully de ined segmen s. The company may ace
ewe compe i o s i ewe compe i o s a e ocused on his ma ke segmen (Ko le ,
2008).
4.4.3. Niche Ma ke ing
Ma ke segmen s a e no mally la ge, iden i iable g oups wi hin a ma ke . Niche
ma ke ing ocuses on subg oups wi hin he segmen s. A niche is a mo e na owly
de ined g oup, usually iden i ied by di iding a segmen in o sub segmen s o by
de ining a g oup wi h a dis inc i e se o who may seek a special combina ion o
bene i s. Whe eas segmen s a e ai ly la ge and no mally a ac se e al compe i o s,
niches a e smalle and no mally a ac only one o a ew compe i o s. Niche ma ke e s
p esumably unde s and hei niches' needs so well ha hei cus ome s willingly pay a
p ice p emium (Ko le , 2008).

32
4.4.4. Mic o Ma ke ing
Segmen and niche ma ke e s ailo hei o e s and ma ke ing p og ams o mee
he needs o a ious ma ke segmen s. A he same ime, howe e , hey do no
cus omize hei o e s o each indi idual cus ome . Thus, segmen ma ke ing and niche
ma ke ing all be ween he ex emes o mass ma ke ing and mic o ma ke ing. Mic o
ma ke ing is he p ac ice o ailo ing p oduc s and ma ke ing p og ams o sui he as es
o speci ic indi iduals and loca ions. Mic o ma ke ing includes local ma ke ing
(in ol es ailo ing b ands and p omo ions o he needs and wan s o local cus ome
g oups—ci ies, neighbo hoods, and e en speci ic s o es) and indi idual ma ke ing
( ailo ing p oduc s and ma ke ing p og ams o he needs and p e e ences o indi idual
cus ome s) (Ko le , 2008).
4.5. Segmen a ion Bases
Wedel and Kamaku a (2000) de ine segmen a ion basis as “a se o a iables o
cha ac e is ics used o assign po en ial cus ome s o homogeneous g oups.” They
classi y segmen a ion bases in o ou ca ego ies, which a e p esen ed in Table 4.4.
Bases
Gene al
P oduc -speci ic
Obse able
Cul u al, geog aphic,
demog aphic and socio-
economic a iables
Use s a us, usage equency,
s o e loyal y and pa onage,
si ua ions
Unobse able
Psychog aphics, alues,
pe sonali y and li e-cycle
Psychog aphics, bene i s,
pe cep ions, elas ici ies,
a ibu es, p e e ences,
in en ion
Table 4.4 – Classi ica ion o Segmen a ion Bases
(Sou ce: Wedel and Kamaku a, 2000)
Gene al segmen a ion bases a e independen o p oduc s, se ices o
ci cums ances and p oduc -speci ic segmen a ion bases a e ela ed o cus ome and
p oduc , se ice and/o pa icula ci cums ances. Fu he mo e, obse able segmen a ion
bases a e measu ed di ec ly and unobse able segmen a ion bases a e in e ed.
33
4.5.1. Obse able Gene al Bases
In ma ke segmen a ion, a widely numbe o bases a e used in his ca ego y, such
us cul u al a iables, geog aphic a iables, neighbo hood, geog aphic mobili y,
demog aphic and socio-economic a iables, pos al code classi ica ions, household li e
cycle, household and company size, s anda d indus ial classi ica ions and
socioeconomic a iables. Also used a e media usage and socioeconomic s a us (Wedel
and Kamaku a, 2000). The obse able gene al bases play an impo an ole in
segmen a ion s udies, whe he simple o complex, and a e used o enhance he
accessibili y o segmen s de i ed by o he bases (Wedel and Kamaku a, 2000).
4.5.2. Obse able P oduc -Speci ic Bases
This kind o segmen a ion bases include a iables ha a e ela ed o buying and
consump ion beha io , like use s a us, usage equency, b and loyal y, s o e loyal y,
s o e pa onage, s age o adop ion and usage si ua ion. These a iables ha e been used
bo h o consume and business ma ke s (Wedel and Kamaku a, 2000).
4.5.3. Unobse able Gene al Bases
Th ee g oups o a iables in his class o segmen a ion bases a e iden i ied: (1)
pe sonali y ai s, (2) pe sonal alues and (3) li es yle.
The i s may include dogma ism, consume ism, locus o con ol, eligion and
cogni i e s yle. The mos equen ly used scale o measu ing gene al aspec s o
pe sonali y in ma ke ing is he Edwa d’s pe sonal schedule.
Rela i ely o he second, he mos impo an ins umen o measu e human
alues and o iden i y alue sys ems is he Rokeach alue su ey.
The hi d is based on h ee componen s: ac i i ies (wo k, hobbies, social e en s,
aca ion, en e ainmen , clubs, communi y, shopping, spo s), in e es s ( amily, home,
job, communi y, ec ea ion, ashion, ood, media, achie emen s) and opinions (o
onesel , social issues, poli ics, business, economics, educa ion, p oduc s, u u e,
cul u e).The li es yle ypology mos used is he VALS sys em, which has been ecen ly
e iewed gi ing i s place o he VALS2 sys em. I is de ined by wo main dimensions:
esou ces (income, educa ion, sel -con idence, heal h, eage ness o buy, in elligence,
e c.) and sel -o ien a ion (p inciple-o ien ed, sel -o ien ed and s a us-o ien ed).These
h ee g oups o a iables a e used almos exclusi ely o consume ma ke s gi ing us “a
34
mo e li elike pic u e o consume s and a be e unde s anding o hei mo i a ions”
(Wedel and Kamaku a, 2000).
4.5.4. Unobse able P oduc -Speci ic Bases
This class o segmen a ion bases comp ises p oduc -speci ic psychog aphics,
p oduc -bene i pe cep ions and impo ance, b and a i udes, p e e ences and beha io al
in en ions. In hese o de , he a iables o m a hie a chy o e ec s, as each a iable is
in luenced by hose p eceding i . Many o hese a iables a e used o consume
ma ke s; howe e hey can also be used o segmen ing business ma ke s (Wedel and
Kamaku a, 2000).
All he segmen a ion bases a e summa ized on Table 4.5 acco ding o he six
c i e ia o e ec i e segmen a ion.
C i e ia
Bases
Iden i iabili y
Sus ainnabili y
Accessibili y
S abili y
Ac ionabili y
Responsi eness
Gene al, Obse able
++
++
++
++
-
-
Speci ic, Obse able
- Pu chase
+
++
-
+
-
+
- Usage
+
++
+
+
-
+
Gene al, Unobse able
- Pe sonali y
+-
-
+-
+-
-
-
- Li es yle
+-
-
+-
+-
-
-
- Psycog aphics
-
+-
+-
-
-
Speci ic, Unobse able
- Psycog aphics
+-
+
-
-
++
+-
- Pe cep ions
+-
+
-
-
+
-
- Bene i s
+
+
-
+
++
++
- In en ions
+
+
-
+-
-
++
++ e y good, + good, +- mode a e, - poo , -- e y poo
Table 4.5 – E alua ion o Segmen a ion Bases
(Adap ed om Wedel and Kamaku a, 2000)
35
4.6. Segmen a ion Me hods
Many segmen a ion me hods a e a ailable and ha e been used. Wedel and
Kamaku a (2000) classi y segmen a ion me hods in wo ways: (1) a p io i o pos hoc;
and (2) desc ip i e o p edic i e.
A segmen a ion app oach is called a p io i when he ype and numbe o
segmen s a e de e mined in ad ance by he esea che and pos hoc when he ype and
numbe o segmen s a e de e mined on he basis o he esul s o da a analysis. (Wedel
and Kamaku a, 2000).
A desc ip i e me hod analyzes he associa ions ac oss a single se o
segmen a ion bases, wi h no dis inc ion be ween dependen o independen a iables.
Such me hod o ms clus e s ha a e homogeneous along a se o obse ed a iables. A
p edic i e me hod analyzes he associa ion be ween wo se s o a iables, whe e one se
consis s o dependen a iables o be explained/ p edic ed by he se o independen
a iables. This me hod o ms clus e s ha a e homogeneous on he es ima ed
ela ionship be ween he wo se s o a iables (Wedel and Kamaku a, 2000).
This classi ica ion o segmen a ion me hods conduc us o ou ca ego ies ha
a e lis ed on Table 4.6.
Me hods
A p io i
Pos hoc
Desc ip i e
Con ingency ables,
Log-linea models
Clus e ing me hods:
Nono e lapping, o e lapping,
Fuzzy echniques, ANN, mix u e
models
P edic i e
C oss- abula ion,
Reg ession, logi and
Disc iminan analysis
AID, CART, Clus e wise
eg ession, ANN, mix u e models
Table 4.6 – Classi ica ion o Segmen a ion Me hods
(Adap ed om Wedel and Kamaku a, 2000)
Despi e his classi ica ion, hyb id o ms o segmen a ion a e also possible and
ha e been applied, combining a p io i and pos hoc app oaches. The hyb id p ocedu e
can be seen as combining he s eng hs o he a p io i and pos hoc app oaches.

PART II
“P ac ice is he bes o all ins uc o s.”
By Publilius Sy us, a Roman w i e
45
5. RESULTS ANALYSIS IN A TECHNICAL
PERSPECTIVE
“Whene e an indi idual o a business
Decides ha success has been a ained,
P og ess s ops.”
By Thomas J. Wa son, an Ame ican scien is
In his sec ion he esul s o clus e ing in a echnical pe spec i e, p o ided by
RapidMine , will be p esen ed and discussed. Ou objec i e he e is o answe he
business p oblem a hand in e ms o he p oduc ion o shi s. We aim, h ough DM
echniques (Clus e ing) and ools (RapidMine ) o de ec and unde s and he ashion
ends on shi s based on he shi s o de s o Bi olino in 2011. Wi h he in o ma ion
esul ing om he clus e ing p ocess
7
, he ashion designe s would be able o iden i y
ashion ends. Gi en ha , we a e going o ocus ou analysis mainly on shi s a ibu es.
5.1. Selec ion o he Clus e ing Resul
Va ious expe iences we e done wi h da a in RapidMine (Annex E) howe e we
had o selec one o hem o analyze in mo e de ail aking in o accoun he pu pose o
ou s udy. The esul selec ed is illus a ed in Figu e 5.1.
7
Due o he lack o space, he de ails om he da a mining p ocess a e p esen ed in Appendix A and
Appendix B.
46
Figu e 5.1 – Resul o K-Medoids Clus e ing Ex ac Clus e P o o ype o k=6
The choice o his esul , wi h a k=6 de ined a p io i, was based on a ious
easons. Be o e enume a ing hem, we mus ell ha his is a e y di icul and
subjec i e ask, which is a p oblem usually associa ed o his clus e ing algo i hm. This
is pa icula ly ue because he au ho s a e no shi designe s. A e doing some
expe imen s wi h di e en numbe o clus e s, we had o do decide o he “bes ” k.
Tha decision was made aking in o accoun he amoun o he da a being clus e ed (a
da ase wi h 10.775 examples and 29 a ibu es), he pu pose o he analysis and he
in o ma ion i could gi e us. Gi en ha , i we had chosen a e y small k, i wouldn’ be
possible o ex ac om he esul s aluable and use ul in o ma ion because i will
emain he e hidden. And i we ha e chosen a e y big k, i was almos impossible o
wo k on and analyze so many da a, despi e he ac ha an inc ease in k educes he
squa ed e o .
A e his he clus e s will be in e p e ed one by one based on he a ibu es ha
a e ele an o p oduc ion and ashion designe s.
5.2. In e p e a ion o Clus e s
Clus e 1
Figu e 5.2 – Rep esen a ion o Clus e 1
47
Clus e 1 g ouped 1.282 shi s o de s which ep esen ed abou 12% o he o al
o de s o he company in 2011. This clus e ells us ha especially young men om he
Uni ed Kingdom (uk) ha e a p e e ence o wo k shi s om Re aile 1 collec ion ha
has he ollowing cha ac e is ics:
A ibu e8
Value
Fab ic
She ield
Fab ic colo
Mul icolo
Fab ic design pa e n
Fine S ipe
Fab ic inish
Easy i on
Fab ic s uc u e
Dobby
Fab ic ma e ial
100% Co on
Colla
Hai Cu away
Colla whi e
no
Cu
Double inc. Cu links
Cu whi e
no
Placke
Folded
Pocke
no
Monog am
no
Hem
Cu ed Hem wi h Gusse s
Back yoke con as
yes
Table 5.1 – Shi s A ibu es Values o Clus e 1
Besides his in o ma ion abou he ab ic cha ac e is ics and he di e en
componen s, i is also impo an o pay a en ion o he physical cha ac e is ics o he
clien s. Fo example, i he clien is e y all o obese, he p oduc ion depa men should
buy mo e quan i y o ha ab ic since i is no enough o know wha he ab ic ype i
needed bu also he quan i y necessa y o p oduce he shi s. This is an impo an issue
because p edic ing ha an iden i ied g oup o clien s wi h a ce ain physical
cha ac e is ics ha e speci ic as es in e ms o shi s and he e o e di e en ashion
8
The desc ip ion o each a ibu e is a ailable on Annex D, on Table D.3, Table D.4 and Table D.5
acco ding o he en i y i desc ibes (cus ome s, shi s, o de s).

48
ends compa ing o o he g oups, will no only enable he p oduc ion depa men o
ha e he necessa y ma e ial bu also o espond in ime o he o de s.
The in o ma ion abou he o de s (shi s and cus ome s a ibu es) is gi en by he
cus ome s o he company h ough i s websi e. Figu e 5.3 and Figu e 5.4 illus a e he
p ocess o designing a cus omized men shi wi h a 3D echnology, whe e he i s
ep esen s shi a ibu es and he second he cus ome a ibu es. Since Bi olino only
p oduces o ul ill he o de s i ecei es, he e is no was e and no s ock.
Figu e 5.3 – Illus a ion o Designing a Men Shi on Bi olino websi e – Shi s
A ibu es
49
Figu e 5.4 – Illus a ion o Designing a Men Shi on Bi olino websi e – Cus ome s
A ibu es
In e ms o he ab ic, he company has abou 300 di e en ypes a ailable o
men shi s. In his clus e he p e e ed ab ic is “She ield” and we can see in Figu e 5.5
ha his ype o ab ic is on he op 13, since in 10.281 men shi s sold by he company
in 2011, 138 we e p oduced wi h his ab ic.
Figu e 5.5 – Fab ics o Men Shi s
0100 200 300 400 500 600 700
G eenwich
london 4
Yo k
Wo ces e
Plymou h
Ha ow
Ma co1
Roco 1
Ba h
london 1
B igh on
Ca di
She ield
Fab ic - Men
Quan i y
50
In e ms o he ab ic colo , he “mul icolo ” is he mos desi ed colo (Figu e
5.6). In a o al o 9.815 o de s o men shi s, 695 we e p oduced wi h his colo .
Company o e s abou 30 di e en ab ic colo s and in 2011 his colo was on op 5.
Figu e 5.6 – Fab ic Colo s o Men Shi s
The ab ic design and pa e n “Fine S ipe”, in he 39 di e en cus ome s’
op ions, was in 2011 on op 11. In 9.815 men shi s o de s, 137 we e p oduced wi h his
ab ic design and pa e n.
Figu e5.7 – Fab ic Design and Pa e n o Men Shi s
0
500
1000
1500
2000
2500
3000
Whi e
Blue & Na y
Blue
Blue & Whi e
Mul icolo
G ey
Black
Pink
Pu ple & Lila
Lilac
Na y Mix
O whi e
Red & Bo deaux
B own
Fab ic Colo - Men
Quan i y
01000 2000 3000 4000 5000
Plain
S ipe
Check
Twill Wea e Plain
Ox o d wea e Plain
P ince o Wales Check
Mini Gingham check
Puppy oo h S ipe
S uc u ed plain
2 colou Bengal S ipe
Fine S ipe
Hai line S ipe
Bengal S ipe
Fab ic Design & Pa e n - Men
Quan i y
51
F om 10 di e en ypes o ab ic inish, he “Easy I on” is undoub edly he s a .
Figu e 5.8 shows ha in almos 50% o men shi s o de s he ype o ab ic inish
applied was his one.
Figu e 5.8 – Fab ic Finish o Men Shi s
Among 9 possible op ions in e ms o shi s ab ic s uc u e, Figu e 5.9
demons a es ha “Dobby” was he hi d mos wan ed.
Figu e 5.9 – Fab ic S uc u e o Men Shi s
The ab ic ma e ial “100% Co on” is he p e e ed by he majo i y o male
cus ome s since in Figu e 5.10 is shown ha mo e han 80% chose his ab ic ma e ial.
7%
47%
24%
0%
4% 2% 16%
1%
0%
1%
Fab ic Finish - Men
Basic
Easy i on
Easy-ca e Oko ex 100
Flannel b ushed
Non i on
Non-i on Oko ex 100
Oko ex 100
P in
0%
10%
5%
6% 1%
5%
1%
55%
0%
17%
Fab ic S uc u e - Men
Co du oy - Velou s
Dobby
End-on-end
He ingbone
Jacqua d
Ox o d
Pinpoin
Poplin
Sa een
Twill
58
A ibu e
Value
Hem
S aigh
Back yoke con as
no
Table 5.5 – Shi s A ibu es Values o Clus e 3
Despi e being e y di e en , his clus e s ill has some a ibu es alues in common wi h
bo h clus e 1 and clus e 2.
A ibu e
Value
Simila i y o
Obse a ions
Fab ic design pa e n
Plain
Clus e 2
See Figu e 5.7
Fab ic inish
Easy i on
Clus e 1 and 2
See Figu e 5.8
Cu
Round Single
Clus e 2
See Figu e 5.12
Placke
Folded
Clus e 1
See Figu e 5.13
Colla whi e
no
Clus e 1 and 2
See Table 5.2
Cu whi e
no
Clus e 1 and 2
See Table 5.2
Table 5.6 – Simila i ies agains Clus e 1 and Clus e 2
This clus e , unlike clus e 1 and clus e 2, ep esen s a small g oup o o de s
wi h some dis inc i e cha ac e is ics. S a ing wi h he ab ic, he “Juan 7”
9
in 10.281
shi s sold by Bi olino in 2011, only 15 shi s we e p oduced wi h his ype o ab ic ye
he ab ic colo “Red & Bo deaux” was he 13 h choice in a anking o 31 di e en
colo s (Figu e 5.6). The ab ic s uc u e “Twill” was he second mos wan ed ype
(Figu e 5.9). Likewise, he ab ic ma e ial “Co on wo old” is he second ype p e e ed
(Figu e 5.10). The ype o colla “Mao” i is no a equen op ion (i is no ep esen ed
in Figu e 5.11) since in o al men shi o de s (9.815) only 31 was his ype o colla .
The “s aigh ” hem was no chosen by he majo i y o cus ome s (Figu e 5.14) and
unlike clus e 1 and clus e 2, clus e 3 ep esen s o de s whe e shi s ha e pocke ,
monog am and do no ha e back yoke con as .
In conclusion, clus e 3 p esen s a e y di e en pa e n in e ms o ashion
ends which is p obably ela ed o, on one hand, a di e en class age (25-34 s. 45-54)
and na ionali y (Uni ed Kingdom s. Ge many) and on he o he hand, o a di e en
con igu a o o he shi used (wo k shi s. ashion shi ) and BMI indices (No mal s.
9
The ab ic “Juan 7” is no ep esen ed in Figu e 5.5 – Fab ic o Men Shi s because o i s li le
ep esen a ion on o e all 302 di e en ypes o ab ic.

59
Obese). La e on his disse a ion, we will be able o see ha he e is e ec i ely a
ela ion be ween all hese ac o s; ha ashion ends o p oduc a ibu es a e no
insepa able om cus ome ’s physical a ibu es.
Clus e 4
Figu e 5.17 – Rep esen a ion o Clus e 4
Simila o he analysis o Clus e 1 and Clus e 2, Clus e 4 ep esen s young
men om he Uni ed Kingdom ha ha e a p e e ence o buying wo k shi s h ough
Re aile 1 websi e. This clus e g ouped 1.570 shi o de s which ep esen ed abou 15%
o he o al company o de s in 2011. The shi s a ibu es ep esen ed in his clus e a e
he ollowing:
A ibu e
Value
Fab ic
London 4
Fab ic colo
Whi e
Fab ic design pa e n
Plain
Fab ic inish
Easy i on
Fab ic s uc u e
Poplin
Fab ic ma e ial
100% Co on
Colla
I alian Semi Sp ead
Colla whi e
no
Cu
Double inc. Cu links
Cu whi e
no
Placke
Folded
Pocke
no
Monog am
no
60
A ibu e
Value
Hem
S aigh
Back yoke con as
yes
Table 5.7 – Shi s A ibu es Values o Clus e 4
Clus e 4 sha es he cha ac e is ics o clus e s 1, 2 and 3 as we can see in Table
5.8.
A ibu e
Value
Simila i y o
Obse a ions
Fab ic colo
Whi e
Clus e 2
See Figu e 5.6
Fab ic design pa e n
Plain
Clus e 2 and 3
See Figu e 5.7
Fab ic inish
Easy i on
Clus e 1, 2 and 3
See Figu e 5.8
Fab ic s uc u e
Poplin
Clus e 2
See Figu e 5.9
Fab ic ma e ial
100% Co on
Clus e 1 and 2
See Figu e 5.10
Colla whi e
no
Clus e 1, 2 and 3
See Table 5.2
Cu
Double inc. Cu links
Clus e 1
See Figu e 5.12
Cu whi e
no
Clus e 1, 2 and 3
See Table 5.2
Placke
Folded
Clus e 1 and 3
See Figu e 5.13
Pocke
no
Clus e 1 and 2
See Table 5.2
Monog am
no
Clus e 1 and 2
See Table 5.2
Hem
S aigh
Clus e 3
See Figu e 5.14
Back yoke con as
yes
Clus e 1 and 2
See Table 5.2
Table 5.8 – Simila i ies agains Clus e 1, Clus e 2 and Clus e 3
As we can see clus e 4 only di e s om he o he h ee clus e s in wo
a ibu es, on he ab ic and on he colla . The ab ic “London 4” was he second mos
common choice (Figu e 5.5) wi h 363 ou o 10.281 men shi s o de ed wi h his ype o
ab ic. The colla ype “I alian Semi Sp ead” is also he second mo e wan ed in 27
di e en ypes, ep esen ing abou 20% o he o al (Figu e 5.11).
In conclusion, clus e 4 is e y simila o clus e 1 and clus e 2 excep in wo
a ibu es. The e o e we can say ha young male wo ke s ha e simila as es and
physical a ibu es wi h some di e ences in e ms o ashion ends and p e e ences.
61
Once again we can con i m ha some common ac o s like age and na ionali y ha e
in luence on as es as well as he pu pose o he buying (business minded o no ).
Clus e 5
Figu e 5.18 – Rep esen a ion o Clus e 5
Clus e 5 g ouped 500 shi s o de s which ep esen ed abou 5% o he o al
o de s o he company in 2011. This clus e is smalle han he o he ou clus e s
analyzed and i ep esen s young men cus ome s om Ge many (de) ha ha e a
p e e ence o ashion shi s om a “Fashion T end” collec ion wi h he ollowing
cha ac e is ics:
A ibu e
Value
Fab ic
Kiwi 9
Fab ic colo
Pu ple & Lila
Fab ic design pa e n
Plain
Fab ic inish
Easy Ca e Oko ex 100
Fab ic s uc u e
Pinpoin
Fab ic ma e ial
100% Co on
Colla
To ino La ge 2 Bu on
Colla whi e
yes
Cu
Round Single
Cu whi e
yes
Placke
Real F on
Pocke
no
Monog am
yes
Hem
S aigh
62
A ibu e
Value
Back yoke con as
no
Table 5.9 – Shi s A ibu es Values o Clus e 5
This clus e app oaches o he o he s clus e s in he ollowing shi s a ibu es:
A ibu e
Value
Simila i y o
Obse a ions
Fab ic design pa e n
Plain
Clus e 2, 3 and 4
See Figu e 5.7
Fab ic ma e ial
100% Co on
Clus e 1, 2 and 4
See Figu e 5.10
Cu
Round Single
Clus e 2 and 3
See Figu e 5.12
Placke
Real F on
Clus e 3
See Figu e 5.13
Pocke
no
Clus e 1, 2 and 4
See Table 5.2
Monog am
yes
Clus e 3
See Table 5.2
Hem
S aigh
Clus e 3 and 4
See Figu e 5.14
Back yoke con as
no
Clus e 3
See Table 5.2
Table 5.10 – Simila i ies agains Clus e 1, Clus e 2, Clus e 3 and Clus e 4
The “Kiwi 9” ab ic was used o p oduce only 7 shi s in he o al o 10.281 men
shi s o de s in 2011. This ype o ab ic is no ep esen ed in Figu e 5.5, only he ones
mo e commonly used. In Figu e 5.6 we can see ha he ab ic colo “Pu ple & Lila” a e
among he mos sough a e and he ab ic inish “Easy Ca e Oko ex 100” was second
on he op-lis o cus ome ’s p e e ences among 8 di e en op ions (Figu e 5.8). The
ab ic s uc u e “Pinpoin ” seems like an unusual choice since i is unde - ep esen ed on
he o e all choices as shown in Figu e 5.9, and in a o al o 27 colla ypes he “To ino
La ge 2 Bu on” is in 6 h place o he lis o cus ome ’s p e e ences (Figu e 5.11). In his
clus e he colla and he cu ha e he same colo as he shi s, an a ibu e alue
(“yes”) ha only is assumed in his clus e .
In conclusion, and speci ically compa ing clus e 5 o clus e 3, we can say ha
despi e he na ionali y (de) and he con igu a o o he shi s ( ashion) being he same,
hese wo clus e s p esen e y di e en ashion ends. This could be pa ially
explained by he age class (25-34 s. 45-54) and o he physical a ibu es such as BMI
indices (no mal s. obese).
63
Clus e 6
Figu e 5.19 – Rep esen a ion o Clus e 6
Clus e 6 g ouped 1.662 shi o de s which ep esen ed abou 15% o he o al
o de s o he company in 2011. This clus e ep esen s young men cus ome s om
Ge many (de) which ha e a p e e ence o ashion shi s om he “Fashion T end”
collec ion wi h he ollowing a ibu es:
A ibu e
Value
Fab ic
Mi o 3
Fab ic colo
Blue & Na y
Fab ic design pa e n
Plain
Fab ic inish
Easy Ca e Oko ex 100
Fab ic s uc u e
He ingbone
Fab ic ma e ial
Co on wo old
Colla
Classic Poin
Colla whi e
no
Cu
Round Single
Cu whi e
no
Placke
Real F on
Pocke
no
Monog am
no
Hem
Cu ed
Back yoke con as
no
Table 5.11 – Shi s A ibu es Values o Clus e 6

64
This clus e has some simila i ies be ween he p eceden s clus e s al eady
analyzed as p esen ed in Table 5.12.
A ibu e
Value
Simila i y o
Obse a ions
Fab ic design pa e n
Plain
Clus e 2, 3,4 and 5
See Figu e 5.7
Fab ic inish
Easy Ca e Oko ex 100
Clus e 5
See Figu e 5.8
Fab ic ma e ial
Co on wo old
Clus e 3
See Figu e 5.10
Colla
Classic Poin
Clus e 2
See Figu e 5.11
Cu
Round Single
Clus e 2, 3 and 5
See Figu e 5.12
Colla whi e
no
Clus e 1, 2, 3, 4
See Table 5.2
Cu whi e
no
Clus e 1, 2, 3, 4
See Table 5.2
Placke
Real F on
Clus e 2 and 5
See Figu e 5.13
Pocke
no
Clus e 1, 2, 4 and 5
See Table 5.2
Monog am
no
Clus e 2 and 4
See Table 5.2
Hem
Cu ed
Clus e 2
See Figu e 5.14
Back yoke con as
no
Clus e 3 and 5
See Table 5.2
Table 5.12 – Simila i ies agains Clus e 1, Clus e 2, Clus e 3, Clus e 4 and Clus e 5
Despi e hose simila i ies, clus e 6 dis inguishes om hem in o he a ibu es
like he ab ic “Mi o 3” which is no a e y equen choice and because o his no
ep esen ed in Figu e 5.5. In o he wo ds, his ype o ab ic was only used on he
p oduc ion o 20 men shi s in a o al o 10.281. The ab ic colo “Blue & Na y” is he
second mos used (Figu e 5.6) and inally, he ab ic s uc u e “He ingbone” is he
ou h choice in a a ie y o 10 di e en ypes (Figu e 5.9).
In conclusion and speci ically compa ing o clus e 3 and clus e 5, due o he
ac hey ep esen cus ome s wi h he same na ionali y (de) and o de s wi h he same
con igu a o ( ashion), clus e 6 shows ha despi e hese common issues we can s ill
ind di e en pa e ns, as es, p e e ences and ashion ends.
65
5.3. Conclusions o he Clus e ing Resul s
In his sec ion we analyzed he esul s o a K-Medoids clus e ing o a k numbe
o clus e s es ablished a p io i (k = 6) based in a da ase o 10.775 examples (n =
10.775) which co espond o he numbe o shi s o de s o Bi olino in 2011.
The analysis was ocused on 29 a ibu es which we e di ided in o h ee
di e en ca ego ies acco ding o he en i y i desc ibes, hey a e he o de s, he shi s
and he cus ome s (Table D.2, Table D.3 and Table D.4 on Annex D). We ocused he
s udy mainly on he a ibu es ha desc ibe he shi s and he o de s because ou
pu pose was o iden i y he shi s ashion ends based on he cus ome s choices in
e ms o shi s a ibu es.
Once he mos common a ibu es alues in shi s o de s we e iden i ied, he
p oduc ion and/o pu chasing depa men and he ashion designe s would be able o
be e pe o m hei asks, because hey al eady ha e in o ma ion abou he ype and
quan i y o he aw ma e ial needed as well he shi s ashion ends.
In sho , we can say ha DM ools and echniques a e indeed aluable
ins umen s o be e unde s and consume as es and p e e ences allowing companies o
be mo e e icien and esponsi e o cus ome ’s eques s and gaining a compe i i e
ad an age.
67
6. RESULTS ANALYSIS IN A MARKETING
PERSPECTIVE
“The g a i ica ion comes in he doing,
No in he esul s.”
By James Dean, an Ame ican ac o
6.1. Iden i ica ion o Segmen s
In o de o pe o m an analysis in a ma ke ing pe spec i e, we ound i use ul o
add mo e a iables
10
(o a ibu es) o he clus e ing p ocess. The new a iables a e
men ioned on Table D.2, Table D.3 and Table D.4 on Annex D and a e ma ked in a
di e en colo .
As a esul o adding mo e a iables o he analysis he esul s a e di e en om
he i s ones, e en hough he numbe o k clus e s and n examples emains he same.
So, o a clus e ing wi h a k=6 de ined a p io i and a da ase o n=10.775 bu now wi h
59 a ibu es and no 29, he new esul s a e as shown in Figu e 6.1.
10
Ins ead o using 29 a ibu es o un he clus e ing, we now ha e 59 a ibu es, i.e., an addi ional 30 han
we ini ially had. Howe e , no all will appea in he new esul because despi e i s impo ance o unning
he clus e ing, hey a e no o much signi ican o he analysis.
74
o he o al o de s o men shi s, ollowed by Ne he lands (nl), Ge many (de), F ance
( ) and Belgium (be).
Figu e 6.3 – Rep esen a ion o Men O de s by Coun y
In e ms o he age, we can see on Figu e 6.4 ha he cus ome s wi h an age
be ween 35 and 45 yea s old a e he cus ome s ha mo e buy Bi olino shi s, ensu ing
almos 30% o he sales.
Figu e 6.4 – Age G oups o Male Cus ome s
Figu e 6.5 shows ha he heigh g oups [170cm-179cm] and [180cm-189cm]
(also ep esen ed in he i s esul s on Figu e 5.1) a e e y ep esen a i e o he
majo i y o male cus ome s and ha he peak o sales is eached p ecisely on he g oup
0
1000
2000
3000
4000
5000
6000
7000
a be ch de dk es lu nl uk
Men O de s by Coun y
Quan i y
0
500
1.000
1.500
2.000
2.500
3.000
16-24 25-34 35-44 45-54 55-64 65-75 >75
Age G oup - Men
Quan i y

75
o heigh s ep esen ed on his segmen ([180cm-189cm]) co esponding o abou 40% o
he sales in 2011.
In e ms o he weigh g oups, simila ly we can see ha he in e als [60kg-
80kg], [80kg-100kg] and [100kg-120kg] a e ep esen ed on bo h esul s (Figu e 5.1 and
Figu e 6.1) and ha he wo i s g oups ensu e abou 74% o he sales (o men shi s)
ela i e o he yea o 2011. This is shown on Figu e 6.6 whe e we can see ha he peak
o sales is a ained o he age g oup ep esen ed on his segmen ([80kg-100kg]).
Figu e 6.5 – Heigh G oups (in cm) o Male Cus ome s
Figu e 6.6 – Weigh G oups (in kg) o Male Cus ome s
0
500
1000
1500
2000
2500
3000
3500
4000
Heigh G oup - Men
Quan i y
0
500
1000
1500
2000
2500
3000
3500
4000
40 - 60 60 - 80 80 - 100 100 - 120 120 - 140 > 140
Weigh G oup - Men
Quan i y
76
S ill compa ing he p e ious esul s (Figu e 5.1) wi h he esul s analyzed in his
chap e (Figu e 6.1) we can no e ha he BMI
11
classes ep esen ed in bo h a e he
same: No mal, O e weigh and Obese (Table D.5 on Annex D). In Figu e 6.7 we can
see ha he mos ep esen a i e class is he No mal (42%) ollowed by he class
O e weigh (33%), ha is ep esen on segmen 1, and he class Obese (20%).
Figu e 6.7 – BMI measu es o Male Cus ome s
On Figu e 6.8 is shown ha he colla g oups mo e ep esen a i e co espond
p ecisely o he wo ex eme measu es, i.e., he colla g oup <36 and he colla g oup
>60. Toge he hey ep esen abou 50% o he o de s o men shi s, whe e 23% e e s
o he colla g oup ep esen ed on his segmen , ha is <36.
11
The BMI (Body Mass Index) is a simple index o weigh - o -heigh ha is commonly used o classi y
unde weigh , o e weigh and obesi y in adul s. I is de ined as he weigh in kilog ams di ided by he
squa e o he heigh in me e s (kg/m2). Sou ce: WHO – Wo ld Heal h O ganiza ion -
h p://apps.who.in /bmi/index.jsp?in oPage=in o_3.h ml
1%
42%
33%
20%
3%
BMIFa ness - Men
Unde weigh
No mal
O e weigh
Obese
Mo bidly Obese
77
Figu e 6.8 – Colla G oups o Male Cus ome s
The colla size, like he BMI, is used as a measu e o he obesi y o he
cus ome s whe e o sizes supe io o 53 he cus ome is conside ed obese (Table D.5 on
Annex D). On Table 6.3 we can see ha only 28% o men cus ome s a e conside ed
obese acco ding o hei colla size.
isColla Obese
Value
Value (%)
Yes
2.711
28%
No
7.104
72%
To al
9.815
100%
Table 6.5 – Obesi y o Male Cus ome s measu ed by he Colla Size
Table 6.6 summa izes he wo men ioned measu es o obesi y
12
whe e we can
see ha only 25% o he male cus ome s a e conside ed obese acco ding o hese
measu es.
12
This emphasis on obesi y measu es o he cus ome s is ela ed o he CoReNe p ojec . CoReNe main
aim is o mee pa icula needs and expec a ions o widely ep esen ed Eu opean consume a ge s - such
as elde ly, obese, disabled, diabe ic people -, ha usually look o clo hes and oo wea wi h pa icula
unc ional equi emen s bu , a he same ime ashionable, high quali y, eco-sus ainable and a an
a o dable p ice (see Annex F o mo e de ails o isi h p://www.co ene -p ojec .eu/node/155).
0
500
1000
1500
2000
2500
3000
< 36
37/38
38/39
39/40
40/41
41/42
42/43
43/44
44/45
45/46
46/47
47/48
48/49
48/50
50/51
51/52
52/53
53/54
54/55
57/58
60>
Colla G oup - Men
Quan i y
78
isObese
Value
Value (%)
Yes
2.438
25%
No
7.377
75%
To al
9.815
100%
Table 6.6 – Obesi y o Male Cus ome s
In conclusion, we can say ha cus ome s in his segmen buy shi s o
p o essional use (p obably mus ollow a o mal business d ess code) and hey do no
seem o be p ice sensi i e, since hey do no use ouche s o ge a discoun on paymen .
Segmen 2
Figu e 6.9 – Rep esen a ion o Segmen 2
Segmen 2 ep esen s women om F ance wi h an age be ween 35 and 44 yea s
old ha buy shi s h ough Bi olino websi e o o he pu poses han p o essional ela ed
(con igu a o ype = pa y shi ). The choices o he cus ome s could no be condi ioned
by a business d ess code bu could be, o ins ance, o a special social e en .
This segmen has an a ibu e ha is dis inc om he o he s, ha is he use o
gi ouche s (Figu e 6.10). In Table 6.7 we can see ha 82% o women used ouche s
on he o de s paymen and gi en ha we can say ha hey a e e y ecep i e o sales
p omo ions and p ice sensi i e.
79
Gende
Vouche s
Nº Vouche s
Nº Vouche s (%)
Women
Yes
786
82%
No
174
18%
To al Women
960
100%
Men
Yes
5.022
51%
No
4.793
49%
To al Men
9.815
100%
To al
Yes
5.808
54%
No
4.967
46%
G and To al
10.775
100%
Table 6.7 – Bi olino Gi Vouche s Usage Analysis
Taking in o accoun ha in 2011, women Bi olino shi s collec ion was o he
i s ime in oduced in o he ma ke , we may commen ha hese ouche s could e e
o a special p omo ion used o s imula e an expe imen based on a new p oduc .
The e o e, his kind o p omo ion could possibly no co espond o a ecu ing se
pa e n. Thus, his could jus be a speci ically a ge ed p omo ion, such as an a emp by
he company o change i s cu en consume base, such as inc easing he womens shi
sales. Conside ing hese esul s i should adop and ou line u he p omo ions and
s a egies o his kind, in an a emp o app oach his public in a be e way.

80
Figu e 6.10 – Bi olino Gi Vouche s
A ibu e
Value
Gende
Women
Coun y
Age G oup
[35-44]
Heigh G oup (cm)
[170-179]
Weigh G oup (kg)
[100-120]
BMIFa ness
Obese
Colla G oup
52/53
isColla Obese
No
isObese
Yes
Table 6.8 – Cus ome s A ibu es Values o Segmen 2
Simila o male cus ome s, emale cus ome s om he Uni ed Kingdom (uk)
lead he Bi olino shi s o de s (Figu e 6.11), howe e in his segmen he women
ep esen ed a e om F ance ( ) ha is he second coun y on he anking ep esen ing
25% o o al o de s o women shi s.
81
Figu e 6.11 – Rep esen a ion o Women O de s by Coun y
Rega ding he age (Figu e 6.12), young women be ween 25 and 34 yea s old a e
he mos ep esen a i e g oup o women cus ome s and i could be he mos p o i able
segmen . No mally hey a e conside ed o be consume is , ashion addic ed, hea y-use s
o ashion p oduc s, en husias ic buye s and much o he ime hey buy on impulse. The
in e al o ages ha comes nex ([35-44]) is he one ep esen ed on his segmen and
ep esen s abou 28% o o al o de s o women shi s.
Figu e 6.12 – Age G oups o Female Cus ome s
0
50
100
150
200
250
300
350
a be ch de es lu nl uk
Women O de s by Coun y
Quan i y
,0
50,0
100,0
150,0
200,0
250,0
300,0
350,0
400,0
16-24 25-34 35-44 45-54 55-64 65-75 >75
Age G oup - Women
Quan i y
82
Women ep esen ed on his segmen ha e a heigh be ween 170cm and 179cm
and, despi e no ep esen ing he majo i y o he emale cus ome s hey ep esen 33%
o he o al o de s o women shi s (Figu e 6.13).
Figu e 6.13 – Heigh G oups (in cm) o Female Cus ome s
Looking a Figu e 6.14 we can conclude ha he weigh o women ep esen ed
on his segmen ([100kg-120kg]) e e s o a mino i y, i.e., only 3% o women (31 in 960
o al o de s) ha bough shi s om Bi olino on 2011 we e o e weigh .
Figu e 6.14 – Weigh G oups (in kg) o Female Cus ome s
0
50
100
150
200
250
300
350
400
450
500
Heigh G oup - Women
Quan i y
0
100
200
300
400
500
600
< 40 40 - 60 60 - 80 80 - 100 100 - 120
Weigh G oup - Women
Quan i y
83
Acco ding o BMI measu es women on his segmen belongs o he obese class,
and on Figu e 6.15 we can see ha obese women we e jus 11%, while he majo i y o
he women cus ome s belong o no mal class, like he men cus ome s.
Figu e 6.15 – BMI measu es o Female Cus ome s
Likewise we can obse e on Figu e 6.16 ha he colla size o women
ep esen ed on his segmen (52/53) e e s o a mino i y (2%), i.e., only 16 o de s in a
o al o 960 e e o his colla size. In e ms o measu ing he obesi y o women
cus ome s based on hei colla size, we can see on Table 6.9 ha only 1% a e
conside ed obese and on Table 6.10 ha , in gene al, obese women a e only 4%.
Figu e 6.16 – Colla G oups o Female Cus ome s
5%
57%
25%
11%
1%
BMIFa ness - Women
Unde weigh
No mal
O e weigh
Obese
Mo bidly Obese
0
20
40
60
80
100
120
140
160
180
< 36 38/39 40/41 42/43 44/45 46/47 48/49 50/51 52/53 54/55
Colla G oup - Women
Quan i y
90
A ibu e
Value
Gende
Men
Coun y
de
Age G oup
[25-34]
Heigh G oup (cm)
[180-189]
Weigh G oup (kg)
[60-80]
BMIFa ness
No mal
Colla G oup
39/40
isColla Obese
No
isObese
No
Table 6.17 – Cus ome s A ibu es Values o Segmen 6
Compa ing his segmen o segmen 5, we no e ha he cus ome s physical
a ibu es a e iden ical, such as he age ([25-34]), he heigh ([180cm-189cm]), he
weigh ([60kg-80kg]) and BMI (No mal), despi e being om di e en coun ies.
Howe e , compa ing segmen 6 o all segmen s, i di e s in he na ionali y (de) and in
he colla size (39/40) (Figu e 6.8).
A ibu e
Value
Simila i y o
Obse a ions
Gende
Men
Segmen 1, 3, 4 and 5
See Table 6.1
Age G oup
[25-34]
Segmen 5
See Figu e 6.4
Heigh G oup (cm)
[180-189]
Segmen 1 and 5
See Figu e 6.5
Weigh G oup (kg)
[60-80]
Segmen 3 and 5
See Figu e 6.6
BMIFa ness
No mal
Segmen 5
See Figu e 6.7
isColla Obese
No
Segmen 1 and 4
See Table 6.5
isObese
No
Segmen 1, 3, 4 and 5
See Table 6.6
Table 6.18 – Simila i ies agains Segmen 1, Segmen 3, Segmen 4 and Segmen 6
In conclusion, we no e a end ela ed o he na ionali y a iable, i.e., we saw ha
cus ome s om a ce ain coun y end o pu chase he shi s h ough an a ilia e o hei
own coun y (Figu e 6.22). Thus, o ins ance, cus ome s om he Uni ed Kingdom (uk)
end o o de he shi s om Re aile 1, which ep esen s abou 75% o he o de s, and
cus ome s om Ge many (de) om Re aile 2, which ep esen s abou 76% o he

91
o de s. This could mean ha cus ome s a e ca e ul when shopping online in he sense
ha hey ha e p e e ence o b ands o s o es ha hey al eady know and us .
Figu e 6.22 – Rela ion be ween Coun y and A ilia e o Male Cus ome s
6.4. Conclusions o he Resul s
The clus e ing esul s (Figu e 6.1) o a numbe o clus e s k=6 de ined a p io i
and wi h mo e a iables han ha used in he i s clus e ing p ocess (Figu e 5.1), has
b ough us, in ac , some addi ional in o ma ion. One example is gi en by segmen 2
ha ep esen s women (Figu e 6.9), since i made possible o he company o segmen
i s consume ma ke on he basis o he gende o he cus ome s, as men and women
ha e di e en expec a ions o ashion p oduc s.
We also ha e seen ha he company could segmen i s ma ke based on he age,
since he changes in body pe o mance and shape ha e a g ea impac on he ashion
choices. On example is illus a ed on Figu e 6.23 whe e we can see ha in e ms o shi
i (le el o shi igh ness o he body) choices, he young cus ome s (segmen s 5 and 6)
p e e he “supe slim i ” and he olde cus ome s he “com o i ”, howe e emaining
he “ egula i ” he mos common choice (Figu e 6.24). O he example is illus a ed on
Figu e 6.25 whe e we can see ha he olde cus ome s (segmen s 1 and 4) a e he ones
0
1000
2000
3000
4000
5000
6000
a be ch de dk es lu nl uk
Re aile 2
Re aile 10
Re aile 1
Re aile 9
Re aile 8
Re aile 7
Re aile 6
Re aile 5
Bi olino
Re aile 4
Re aile 3
Coun y and A ilia e
92
ha like o use pocke on shi s he mos , especially he ones o he ype “mi ed”
(Figu e 6.26).
Figu e 6.23 – Rela ion be ween Fi and Age o Male Cus ome s
Figu e 6.24 – Fi choices o Men Shi s
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
16/24 25/34 35/44 45/54 55/64 65/75 > 75
Supe Slim Fi
Regula
Com o Fi
Fi and Age (Men)
21%
57%
23%
Fi - Men
Com o i
Regula
Supe slim i
93
Figu e 6.25 – Rela ion be ween Pocke and Age o Male Cus ome s
Figu e 6.26 – Type o Pocke o Male Cus ome s
The mo e a iables we add he mo e possibili ies o segmen ing he ma ke .
Thus, he company could also segmen he ma ke geog aphically by coun y o locally
(pos al code), since cos ume s om di e en na ionali ies ha e di e en equi emen s
o ashion and clo hing p oduc s, and much o he ime hei choices a e in luenced by
social and cul u al alues. Figu e 6.27 shows he ep esen a ion o he cus ome ’s
coun ies in e ms o company o de s in 2011 and we can see ha he Uni ed Kingdom
(uk) leads. In Figu e 6.28 is shown ha he a ilia es (Bi olino, Re aile 1, Re aile 2)
h ough which he company makes much o i s sales co esponds o he na ionali y o
some o hei bes cus ome s (Belgium, Uni ed Kingdom, Ge many).
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Flap wo
pocke s
o mal mi ed ound single
lap
pocke
s aigh
45/54
35/44
25/34
Pocke and Age G oup - Men
3%
1%
79%
8%
3%
7%
Pocke - Men
Flap wo pocke s
Fo mal
Mil ed
Round
Single Flap Pocke
S aigh
94
Figu e 6.27 – To al O de s by Coun y o Bi olino Shi s in 2011
Figu e 6.28 – To al O de s by A ilia e o Bi olino Shi s in 2011
Ano he al e na i e o segmen ing he Bi olino cos ume s ma ke is making i
on he basis o he pu pose o in en ion o buying acco ding o, o ins ance, he
ca ego ies “wo k shi ”, “ ashion shi ” and “pa y shi ” (con igu a o o collec ion
ype). We can hen in e , o example, ha hey buy mo i a ed by p o essional
equi emen s (e.g. segmen s 1, 4 and 5), in e es on ashion (e.g. segmen 6), o by
social e en s equi emen s (e.g. segmen s 2 and 3).
We ha e also iden i ied ano he segmen a ion a iable ha is he p ice
sensi i i y, which could be measu ed by he usage o gi ouche s o ge a discoun on
paymen . We concluded ha women a e mo e p ice sensi i e han men gi en ha mo e
000%
010%
020%
030%
040%
050%
060%
Nº o O de s by Coun y
Nº o O de s (%)
000%
010%
020%
030%
040%
050%
060%
Nº o O de s by A ilia e
Nº o O de s (%)
95
han 82% used a ouche (Table 6.7) on shi s paymen . Women a e gene ally mo e
ecep i e o p omo ions o his ype and mo e open o expe imen ing new p oduc s han
men. Howe e we do no know i i was ela ed o a speci ic p omo ion wi h he
in en ion o allowing he expe imen o a new p oduc , howe e i yes he company
should s udy o he ways o a ac he emale public and imp o e sales.
In sho , in his chap e we analyzed he p o ile o Bi olino cus ome s and hei
ela ion o he p oduc (shi s) based on some segmen a ion a iables ha we e
iden i ied. We concluded ha hei choices in e ms o shi s a ibu es a e g ea ly
in luenced and condi ioned by nume ous ac o s such as, hei physical a ibu es, age,
gende , na ionali y (coun y), pu pose o buying. Gi en ha , he company mus pay
a en ion o hose di e ences and adjus i s p oduc and ma ke ing s a egies o he
di e en segmen s iden i ied.

97
7. CONCLUSION
“I 's mo e un o a i e a conclusion
Than o jus i y i .”
By Malcolm Fo bes, an Ame ican publishe
7.1. Summa y
The aim o his wo k was o show he p ac ical applicabili y o Da a Mining
(DM) echniques and ools o ma ke segmen a ion in he con ex o he cus omized
ashion indus y. The s udy was based on da a p o ided by Bi olino, a Belgian
company ha p oduces and sells cus om ailo ed shi s, ela i e o i s shi s o de s o
2011.
We s a ed he wo k by doing some li e a u e e iew abou he subjec s co e ed.
We aimed o gi e an o e iew abou DM (de ini ion, asks, p ocess, and me hodology),
Clus e ing (de ini ion, goals, s ages, algo i hms and alida ion) and Segmen a ion in
Ma ke ing (de ini ion, e ec i eness, p ocess, le els, bases, me hods and me hodology).
We hen desc ibed wha we did in p ac ice based on hose heo e ical insigh s
and he da a a ailable. We ollowed he s eps o CRISP-DM (business unde s anding,
da a unde s anding, da a p epa a ion, modeling, e alua ion and deploymen ), a gene al
me hodology o suppo he DM p ocess. The modeling s ep was ca ied ou using he
DM so wa e RapidMine . In e ms o he segmen a ion me hodology, we adop ed a
non-o e lapping (each subjec belongs o a single segmen only) non-hie a chical (s a
om a andom ini ial di ision o he subjec s in o a p ede e mined numbe o clus e s,
and eassign subjec s o clus e s un il a ce ain c i e ion is op imized) clus e ing me hod
called K-Medoids. As we know, clus e ing me hods a e commonly used in ma ke ing
o he iden i ica ion and de ini ion o ma ke segmen s ha help companies ocus hei
ma ke ing s a egies. Then, we p oceeded wi h he ma ke segmen a ion and iden i ied
possible segmen a ion bases acco ding o he classic segmen a ion a iables o
consume ma ke s (demog aphic, geog aphic, psychog aphic and beha io is ic)
sugges ed by Ko le and complemen ed by Wedel and Kamaku a.
98
In his s udy we un he clus e ing p ocess wice o a da ase o n = 10.775 shi s
o de s and a numbe o clus e s k = 6 de ined a p io i, wi h he di e ence on he
numbe o a ibu es (desc ibing he o de s, shi s and cus ome s) used. On he i s
clus e ing p ocess we used 29 a ibu es and on he second we added 30 mo e. The
choice o he a ibu es o use in each p ocess depended on he scope o he analysis, i.e.,
he i s analysis was based on a echnical pe spec i e (p oduc ion and design o he
shi s) and he second on a ma ke ing pe spec i e (segmen a ion o shi s cus ome s). A
he end, we we e able o see how esul s om a clus e ing p ocess could di e simply
changing he numbe o he a ibu es.
Gi en he esul s we s a ed by analyzing he i s esul s mainly ocusing on he
a ibu es ha desc ibe he shi s a ibu es ( ab ic colo , ma e ial, design and pa e n,
inish, s uc u e, e c.) wi h he pu pose o inding some pa e ns in he da a ha could be
ans o med in o use ul in o ma ion o suppo he company’s decisions in e ms o
p oduc ion and design. Then we analyzed he second esul mainly ocusing on he
a ibu es ha desc ibe he cus ome s (gende , age, coun y, heigh , weigh , BMI, colla
size, e c.) in an a emp o di ide he ma ke o Bi olino shi s cus ome s in o dis inc
g oups o buye s who ha e di e en needs, cha ac e is ics, o beha io s, and who migh
equi e sepa a e p oduc s o ma ke ing p og ams.
The analysis o bo h esul s allows us o conclude ha DM echniques and ools
a e indeed e y use ul when analyzing as quan i ies o da a. The i s clus e ing esul
allowed us o answe one o he aspec s om ou business p oblem, i.e., how could he
ashion designe s iden i y and p edic he ashion p o iles and ends. Simila ly, he
second clus e ing esul s enabled us o espond o ano he aspec o ou business
p oblem ha is how he company could segmen i s ma ke in di e en g oups o
cus ome s wi h dis inc cha ac e is ics, needs, p e e ences, as es and beha io s and hen
adap o adjus i s s a egies o espond mo e e ec i ely o hei equi emen s. I will
also allow he company o iden i y which segmen s a e mos p o i able and ocus i s
e o s in ha di ec ion.
99
7.2. Recommenda ions
This s udy ga e us new insigh s abou he ashion indus y. Thus, based on he
da a abou Bi olino cus ome s da abase and on he esul s ob ained om clus e ing and
segmen ing he da a, we ecommend:
 The company should s eng hen i s e o s o expand in o he women’s clo hing
ma ke , since hey ep esen a ma ke segmen ha is e y a en i e o ashion
issues and ecep i e o sales p omo ions and o expe imen ing new p oduc s.
Gi ouche s we e he s a egy adop ed so a by Bi olino o add ess his new
segmen . Gi en he posi i e esponse o his impo an segmen , he company
should conside adop ing p omo ions o di e en ypes.
 As he olde -age consume ma ke is inc easing, he company should also
expand in ha di ec ion. They should be inclusi e in hei ma ke ing and
p omo ional campaigns, highligh ing he bene i s o ashion p oduc s as olde
consume s conside hei clo hing consump ion in a mo e s a egic manne han
younge consume s. Unde s anding he a i udes and beha io o he ma u e
consume s will, on one hand, p o ide guidance o p oduc de elopmen and
design and, on he o he hand, c ea e oppo uni ies and challenges o he
ashion business o ocus on his inc easingly impo an ma ke .
 The company should be awa e o he implica ions o he age in ashion
consume choices, since consume s o all ages psychologically need o exp ess
hei indi idual s yle and as e h ough a choice o be e quali y p oduc s.
Designe s should add alue ia ashion p oduc s ha anscend all ages. They
should pa icula ly p omo e he inclusion aspec o he p ocess o buying
clo hes, no only in he case o olde cus ome s bu also in he case o cus ome s
wi h some kind o disabili y o pa hology.
 Beyond he age and physical condi ions, o he ac o s ha e much in luence on
cus ome s ashion choices and he company mus ha e o ake i in o accoun
when designing he shi s. They a e, o ins ance, he gende (men and women
ha e di e en expec a ions o ashion p oduc s), he na ionali y (cus ome s om
di e en coun ies ha e in luence o social and cul u al alues wi h which hey
iden i y), he pu pose o he buying ( o example, p o essional), e c.
106
includes able, eco d, a ibu e selec ion, ans o ma ion and ‘cleaning’
16
o da a. The
ini ial da ase p o ided by Bi olino was a anged in o de o be modeled on
RapidMine
17
(e.g., some missing a iables we e elimina ed, because hey we e no
signi ican no in quan i y no in quali y o he s udy; some a iables we e ans o med
in a di e en ype, because o he algo i hm used hus equi ed). The e o e, no much
e o was equi ed o his ask.
A ibu e
A ibu e
Class
Finding
Gende
Cus ome
91% o cus ome s a e men (9.815) and 9% a e women (960).
Coun y
Cus ome
57% o cus ome s a e om he Uni ed Kingdom (uk).
Heigh (cm)
Cus ome
68% o cus ome s belong o in e al o heigh [180-189], [170-
179].
Weigh (kg)
Cus ome
73% o cus ome s belong o in e al o weigh [60-80], [80-
100].
Age (yea s)
Cus ome
76% o cus ome s belong o in e al o age [35-44], [25-34],
[45-54].
isObese
(weigh )
Cus ome
77% o cus ome s a e no obese (weigh <100Kg) and 23%
obese (weigh >100kg).
A ilia es
O de
88% o o al o de s we e made h ough Bi olino and
Re aile 118 (48% and 40%, espec i ely).
Con igu a o
ype
O de
The con igu a o ype p e e ed by women is “Bespoke
Women” (75%) and by men is “Wo k Shi ” (35%) ollowed
by “Bespoke Shi ” (22%) and “Fashion Shi ” (20%).
Colla G oup
Shi
21% o shi s ha e he smalles colla (<36) and 25% he
bigges colla (>60).
isColla Obese
Shi
23% o shi s a e o obese pe sons and 77% no .
Collec ion
Type
Shi
The collec ion ype p e e ed by women is “Cha ming” (56%)
ollowed by “Simplissime” (11%).
Colla Whi e
Shi
Only 9% o shi colla s a e whi e, i.e., a e no he same colo
16
Da a cleansing is he p ocess o ensu ing ha all alues in a da ase a e consis en and co ec ly
eco ded.
17
See Annex C o mo e de ails abou RapidMine so wa e o isi h p:// apid-i.com/.
18
Fo con iden iali y easons, we canno disclose he names o he e aile s.

107
A ibu e
A ibu e
Class
Finding
as shi s.
Has
Monog am
Shi
Only 27% o shi s ha e monog ams.
Has Pocke
Shi
Only 45% o cus ome s like shi s wi h pocke s.
Table A.1 – Findings om Explo a o y Da a Analysis
D. Modeling
A e he p e ious phases we e comple ed, i was ime o selec and apply a
modeling echnique. The echnique chosen as being he mos app op ia e o ou DM
p oblem ype, was he Clus e ing, mo e speci ically, he K-Medoids algo i hm, and
gi en he ype o a iables a hand (binominal and polynominal). This echnique
suppo s nominal a iables ( he ypes o a iables suppo ed by RapidMine a e
desc ibed on Table C.1 in Annex C), while o he me hods, such as he K-Means
algo i hm, can only deal wi h nume ical da a.
E. E alua ion
An essen ial s ep o he success o a DM p ojec is he ca e ul e alua ion o he
model buil , be o e p oceeding o i s inal deploymen . I is impo an o e alua e i and
e iew he s eps done o i s c ea ion, o be ce ain ha he model p ope ly achie es he
business objec i es. A he end o his phase, a decision on he use ulness o he DM
esul s should be eached. In his s udy, we ied o e alua e he clus e ing esul s on
RapidMine , howe e we we e no capable o in e p e ing he esul s o such a measu e,
as will be discussed la e .
F. Deploymen
When he model expec ed o be able o achie e he business goals, he
knowledge ob ained has o be o ganized and p esen ed in a way ha can be
ope a ionalized. The deploymen could simply consis o a gene a ion o a epo o in a
implemen a ion o he DM p ocess ac oss he company. Ne e heless, wha is impo an
he e is ha he cus ome unde s ands wha ac ions need o be ca ied ou in o de o
ac ually make use o he model c ea ed.
109
APPENDIX B
Clus e ing using RapidMine
He e we will desc ibe he s eps o he clus e ing p ocess on RapidMine wi h
some illus a ions.
1. Reposi o y
A e p epa ing he da ase o modeling, we we e able o impo i o he
RapidMine Reposi o y
19
and named i as “Fi s Da a Se ”. The da a able
20
is
composed o 29 a ibu es (o a iables), 10.775 examples (numbe o Bi olino shi
o de s) and he alues ha hey could assume. This able ep esen s he so-called Me a
da a which p o ides in o ma ion abou he da a. I includes de ails such as he numbe
and ype o da a s o ed, whe e i is loca ed, how i is o ganized, and so on (Delma e and
Hancock, 2001). This is ypically much less oluminous han he da a i sel and gi es
he analys an excellen idea o which cha ac e is ics a pa icula da ase has. In a ce ain
sense, he Me a da a is he wa ehouse a chi ec u e, because i p o ides he subs ance
upon which all access and applica ions a e based (Delma e and Hancock, 2001).
Figu e B.1 – Reposi o y View on RapidMine
19
The Reposi o y se es as a cen al s o age loca ion o da a and analysis p ocesses. (Sou ce:
RapidMine Use Manual)
20
The Da a Table is p esen ed in Annex D, Table D.1.
110
Figu e B.2 – Da a Table “Fi s Da a Se ”
2. P ocess
In o de o s a he p ocess, we i s ha e o selec he ope a o Reposi o y
Access – Re ie e o ge access o he da a able. Then we selec he ope a o Modeling –
Clus e ing and Segmen a ion – K-Medoids and un he p ocess bu be o e unning he
p ocess we ha e o de ine k, i.e., he numbe o clus e s and also he numbe o uns and
he measu e ype. We selec ed Mixed Measu es, mo e p ecisely he Mixed Euclidean
Dis ance, because his is he dis ance me ic mos widely used o his algo i hm. The
p ocess is hen eady o un and since i is an in ensi e ask, he esul s ake some ime
o be eady o display.
A e his, he esul iew didn´ gi e us he clus e s p o o ype, i.e., he alues
ha each a ibu e will likely assume so we ha e also o selec he ope a o Modeling –
Clus e ing and Segmen a ion – Ex ac Clus e P o o ypes. The ope a o Ex ac Clus e
P o o ype gene a es an ExampleSe consis ing o he Clus e P o o ypes. Fla clus e
algo i hms like K-Means o K-Medoids clus e he da a a ound some p o o ypical da a
ec o s. Fo example, K-Means uses he cen oid o all examples o a clus e . This
ope a o now ex ac s hese p o o ypes and s o es hem in an ExampleSe o u he
111
use, equi ing he inpu a cen oid Clus e Model, which is p ecisely he ou pu o he
clus e ing.
We did some expe imen s o di e en alues o k numbe o clus e s and used
some da ase samples. In addi ion o hese expe imen s, we used he ope a o Da a
T ans o ma ion – A ibu e Se Reduc ion and T ans o ma ion – Selec ion – Selec
A ibu es and de ined, o di e en a ibu es such as a ilia es, BMI, e c., alues om
which he clus e ing would be based in. Some examples o hese expe imen s a e shown
in Annex E.
Figu e B.3 – P ocess View on RapidMine
3. Resul s
A e unning he p ocess, he esul s iew is a ailable and we a e able o see he
ExampleSe in di e en ways. In he case o he Clus e ing i sel , some o he a ailable
iews a e “Tex View”, “Cen oid Table” and “Cen oid Plo View”. In he case o he
ex ac ion o he espec i e clus e p o o ypes, some o he esul iews a ailable a e
“Me a Da a View”, “Da a View” and “Plo View”, and e en “Ad anced Cha s”.

112
Figu e B.4 – Resul s View (“Tex View”) on RapidMine
In he “Tex View” we can see he numbe o he clus e s and hei size which is
called he Clus e Model.
Clus e
Size
Size (%)
1
1.282
12%
2
4.889
45%
3
872
8%
4
1.570
15%
5
500
5%
6
1.662
15%
To al
10.775
100%
113
Figu e B.5 – Resul s View (“Cen oid Table”) on RapidMine
RapidMine assigns a di e en numbe acco ding o he di e en alues ha
each a ibu e could possibly assume. Fo example, i assigns alue “0” o a ibu e
“gende ” whe e ze o ac ually means “men” and alue “1” o “women”, and so on.
114
Figu e B.6 – Resul s View (“Cen oid Plo ”) on RapidMine
In cen oid plo , he philosophy is he same howe e he isual p esen a ion is
g aphical and no abula ed.
115
Figu e B.7 – Resul s View (“Me a Da a”) on RapidMine
In da a iew (Figu e B.8) we can see he clus e p o o ypes, in o ma ion ha
complemen s he Clus e Model (Figu e B.4).
Figu e B.9 (“Plo View”) shows one o he nume ous possibili ies o
isualiza ion o he da ase a ailable on RapidMine .
122
ecommenda ions o imp o emen ”, Aus alasian Jou nal o Ma ke Resea ch,
11(2), pp. 5-12.
DOLNICAR, S., and LEISCH, F. (2003), “Da a-D i en Ma ke Segmen a ion – A
S uc u e-Based Concep ual F amewo k o Managemen Decision Suppo ”,
Aus alian and New Zealand Managemen Academy Con e ence, Adelaide, Sou h
Aus alia, 1-3 Decembe .
FAYYAD, U. and UTHURUSAMY, R. (2002), “E ol ing Da a Mining in o Solu ions
o Insigh s”, Communica ions o he ACM, Vol. 45, Nº 8.
HAND, D., MANNILA, H. and SMITH, P. (2001), P inciples o Da a Mining, The MIT
P ess, Massachuse s Ins i u e o Technology.
HALKIDI, M., BATISTAKIS, Y. and VAZIRGIANNIS, M. (2001), “On Clus e ing
Valida ion Techniques”, Jou nal o In elligen In o ma ion Sys ems, Vol. 17:2/3,
pp. 107-145, Kluwe academic Publishe s.
HALKIDI, M., BATISTAKIS, Y. and VAZIRGIANNIS, M. (2002), “Clus e Validi y
Me hods: Pa I”, SIGMOND Reco d, Vol. 31, Nº 2.
HARDING, J.A., SHAHBAZ, M., SRINIVAS and KUSIAK, A. (2006), “Da a Mining
in Manu ac u ing: A Re iew”, Jou nal o Manu ac u ing Science and
Enginee ing, Vol. 128.
HASTIE, T., TIBSHIRANI, R. and FRIEDMAN, J. (2001), The Elemen s o S a is ical
Lea ning: Da a Mining, In e ence, and P edic ion, Sp inge -Ve lag New Yo k,
Inc.
HSU, C. (2009), “Da a mining o imp o e indus ial s anda ds and enhance p oduc ion
and ma ke ing: an empi ical s udy in appa el indus y”, Expe Sys ems wi h
Applica ions, Nº36, pp. 4185-4191, Else ie L d.
JAIN, A.K., MURTY, M.N. and FLYNN, P.J. (2000), “Da a Clus e ing: A Re iew”,
ACM Compu ing Su eys, Vol. 31, Nº 3.
JAIN, A. K. (2009), “Da a clus e ing: 50 yea s beyond K-Means”, Pa e n Recogni ion
Le e s, Else ie B.V.
KANTARDZIC, M. (2003), Da a Mining - Concep s, Models, Me hods, and
Algo i hms, John Wiley & Sons, INC.

123
KAU, A., TANG, Y. and GHOSE, S. (2003), “Typology o online shoppe s”, Jou nal
o Consume Ma ke ing, Vol. 20, nº 2, pp. 139-156, MCB UP Limi ed.
KETTENRING, J. (2009), “A pa en analysis o clus e analysis”, Applied S ochas ic
Models in Business and Indus y, Nº25, pp. 460-467.
KIM, K. and AHN, H. (2008), “A ecommende sys em using GA K-Means clus e ing
in an online shopping ma ke ”, Expe Sys ems wi h Applica ions, Nº34, pp. 1200-
1209, Else ie L d.
KO, E., KIM, E., TAYLOR, C., KIM, K. and KANG, I. (2007), “C oss-na ional ma ke
segmen a ion in he ashion indus y: A s udy o Eu opean, Ko ean, and US
consume s”, In e na ional Ma ke ing Re iew, Vol. 24, Nº 5, pp. 629-651, Eme ald
G oup Publishing Limi ed.
KOTLER, P. and KELLER, K. (2012), A F amewo k o Ma ke ing Managemen , 5 h
ed. In e na ional ed. - Bos on: Pea son.
KOTLER, P. and ARMSTRONG, G. (2012), P inciples o Ma ke ing, 14 h edi ion,
Pea son Educa ion, Inc.
KOTLER, P. and KELLER, K. (2012), Ma ke ing Managemen , 14 h ed. global
ed. Ha low: Pea son.
KOTLER, P. and KELLER, K. (2008), Ma ke ing Managemen , 13 h edi ion, P en ice
Hall.
KOTLER, P., WONG, V., SAUNDERS, J. and ARMSTRONG, G. (2005), P inciples
o Ma ke ing, 4 h Eu opean Edi ion, Pea son Educa ion L d.
KOTLER, P., RACKHAM, N. and KRISHNASWAMY, S. (2006), “Ending he Wa
be ween Sales and Ma ke ing”, Ha a d Business Re iew.
KOTLER, P. (1999), “Ko le on Ma ke ing: How o C ea e, Win, and Domina e
Ma ke s”, USA: The F ee P ess (FP).
KOZAR, J. and DAMHORST, M. (2008), “Olde women’s esponses o cu en ashion
models”, Jou nal o Fashion Ma ke ing and Managemen , Vol. 12, Nº 3, pp. 338-
350, Eme ald G oup Publishing limi ed.
LAROSE, D. (2005), Disco e ing Knowledge in Da a: An In oduc ion o Da a
Mining, John Wiley & Sons, Inc.
124
LEE, S. J. and SIAU, K. (2001), “A e iew o da a mining echniques”, Indus ial
Managemen & Da a Sys ems, Vol. 101, Nº 1, pp. 41-46, MCB Uni e si y P ess.
LEGÁNY, C., JUHÁSZ, S. and BABOS, A. (2006), 5 h WSEAS In . Con . on A i icial
In elligence, Knowledge Enginee ing and Da a Bases, Mad id, Spain, Feb ua y
15-17, pp. 388-393.
LEVINE, EREL (1999), “Un-Supe ised Es ima ion o Clus e Validi y – Me hods and
Applica ions”, MSc Thesis, Feinbe g G adua e School, Weizmann Ins i u e o
Science, New Yo k, Uni ed S a es o Ame ica.
LIM, J., CURRIM, I. and ANDREWS, R. (2005), “Consume he e ogenei y in he
longe - e m e ec s o p ice p omo ions”, In e na ional Jou nal Resea ch in
Ma ke ing”, Nº 22, pp. 441-457, Else ie B.V.
LO, W., HONG, T. and JENG, R. (2008), “A amewo k o E-SCM mul i-agen
sys ems in he ashion indus y”, In e na ional Jou nal o P oduc ion Economics,
Nº114, pp. 594–614, Else ie B.V.
MACCARTY, J. and HASTAK, M. (2007), “Segmen a ion app oaches in da a-mining:
A compa ison o RFM, CHAID, and logis ic eg ession”, Jou nal o Business
Resea ch, Nº60, pp. 656-662, Else ie inc.
MIRKIN, B. (2005), Da a Clus e ing o Da a Mining: A Da a Reco e y App oach,
Boca Ra on: Chapman and Hall.
MOOI, E. and M. SARSTEDT (2011), A Concise Guide o Ma ke Resea ch, Sp inge -
Ve lag Be lin Heidelbe g.
MOORE, M. and FAIRHURST, A. (2003), “Ma ke ing capabili ies and i m
pe o mance in ashion e ailing”, Jou nal o Fashion Ma ke ing and
Managemen , Vol. 7, Nº 4, pp. 386-397, MCB UP Limi ed.
NEWBERY, M., MEULEN, K. (2010), “Tomo ow’s clo hing e ail: sec o s, ma ke s
and ou es”, Uni ed Kingdom: A oq L d.
NGAI, E.W.T., XIU, L. and CHAU, D.C.K. (2009), “Applica ion o da a mining
echniques in cus ome ela ionship managemen : A li e a u e e iew and
classi ica ion”, Expe Sys ems wi h Applica ions, Nº36, pp. 2592–2602, Else ie
L d.
125
O’CASS, A. (2004), “Fashion clo hing consump ion: an eceden s and consequences o
ashion clo hing in ol emen ”, Eu opean Jou nal o Ma ke ing, Vol. 38, Nº 7, pp.
869-882, Eme ald G oup Publishing Limi ed.
PARK, H. and JUN, C. (2009), “A simple and as algo i hm o K-Medoids
clus e ing”, Expe Sys ems wi h Applica ion,s Nº36, pp. 3336–3341, Else ie L d.
PARK, JUNG HYUN (2009), “Technical Conside a ions o he Design o Sma
Appa el o he O e weigh ”, Thesis o he Deg ee o Mas e o Science, Facul y
o No h Ca olina S a e Uni e si y, Raleigh, Uni ed S a es o Ame ica.
PASCUAL, D., PLA, F. and SÁNCHEZ, J. (2010), “Clus e alida ion using
in o ma ion s abili y measu es”, Nº 31, pp. 454-461, Else ie B.V.
PEDRYCZ, W. (2005), Knowledge-Based Clus e ing: F om Da a o In o ma ion
G anules, New Je sey: John Wiley & Sons, Inc.
PHAM, D. T., DIMOV, S. S. and C. D. NGUYEN (2005), “Selec ion o K in K-Means
clus e ing”, Jou nal o Mechanical Enginee ing Science, Vol. 219, Pa C,
IMechE.
PHAU, I., and LO, C. (2004), “P o iling ashion inno a o s: A s udy o sel -concep ,
impulsi e buying and In e ne pu chase in en ”, Jou nal o Fashion Ma ke ing
and Managemen , Vol. 8, Nº 4, pp. 399-411, Eme ald G oup Publishing Limi ed.
PRIEST, A. (2005), “Uni o mi y and di e en ia ion in ashion”, In e na ional Jou nal
o Clo hing Science and Technology, Vol. 17, Nº 3/4, pp. 253-263, Eme ald
G oup Publishing Limi ed.
RENDÓN, E., ABUNDEZ, I., ARIZMENDI, A. and QUIROZ, E. (2011), “In e nal
e sus Ex e nal clus e alida ion indexes”, In e na ional Jou nal o Compu e s
and Communica ions, Issue 1, Vol. 5.
RYGIELSKI, C., WANG, J. and YEN, D. (2002), “Da a Mining echniques o
cus ome ela ionship managemen ”, Technology in Socie y, Nº24, pp. 483-502,
Else ie Science L d.
ROCHA, M., HAMMOND, L., HAWKINS, D. (2005), “Age, gende and na ional
ac o s in ashion consump ion”, Jou nal o Fashion Ma ke ing and Managemen ,
Vol. 9, Nº4, pp. 390-390, Eme ald G ouping Publishing Limi ed.
126
RUD, O. (2001), Da a Mining Cookbook: modeling da a o ma ke ing, isk and
cus ome ela ionship managemen , New Yo k: John Wiley & Sons, Inc.
SCHAFER, J., KONSTAN, J. and RIEDL, J. (2001), “E-comme ce Recommenda ion
Applica ions”, Da a Mining and Knowledge Disco e y, Nº 5, pp. 115-153,
Kluwe Academic Publishe s.
SCOTT, D. (2007), The New Rules o Ma ke ing & PR, New Yo k: John Wiley & Sons,
Inc.
SHAW, M., SUBRAMANIAM, C., TAN, G. and WELGE, M. (2001), “Knowledge
managemen and da a mining o ma ke ing”, Decision Suppo Sys ems, Nº31,
pp. 127-137, Else ie Science B.V.
SIDDIQUI, N., O’MALLEY, A., MCCOLL, J. and BIRTWISTLE, G. (2003), “Re aile
and consume pe cep ions o online ashion e aile s: Web si e design issues”,
Jou nal o Fashion Ma ke ing and Managemen , Vol. 7, Nº 4, pp. 345-355, MCB
UP Limi ed.
SINGH, S. and CHAUCHAN, N. C. (2011), “K-Means s. K-Medoids: A Compa a i e
S udy”, Na ional Con e ence on Recen T ends in Enginee ing & Technology,
B.VM. Enginee ing College, V. V. Naga , Guja a , India.
SIVOGOLOVKO, E. (2012), “Valida ing clus e s uc u es in Da a Mining asks”,
EDBT/ICDT Wo kshops Ma ch 26-30, Be lin, Ge many, ACM.
SMITH, W. (1956), “P oduc di e en ia ion and ma ke segmen a ion as al e na i e
ma ke ing s a egies”, Jou nal o Ma ke ing, 21, 3-8.
SU, Z. and XU, D. (2011), “Li es yle o Obese Popula ion and B and Choosing o
Pe sonalized Clo hing”, IEEE.
TAN, P., STEINBACH, M. and KUMAR, V. (2006), In oduc ion o Da a Mining:
Ins uc o ’s Solu ion Manual, Pea son Addison-Wesley.
TRINH, G., DAWES, J. and LOCKSHIN, L. (2009), “Do p oduc a ian s appeal o
di e en segmen s o buye s wi hin a ca ego y?”, Jou nal o P oduc & B and
Managemen , 18/2, pp. 95-105, Eme ald G oup Publishing Limi ed.
127
TURQUIN, E., WITHER, J., BOISSIEUX, L., GANI, M. and HUGHES, J. (2007), “A
Ske ch-Based In e ace o Clo hing Vi ual Cha ac e s”, IEEE Compu e
G aphics and Applica ions.
VELMURUGAN, T. and SANTHANAM, T. (2010), “Compu a ional Complexi y
be ween K-Means and K-Medoids Clus e ing Algo i hms o No mal and
Uni o m Dis ibu ions o Da a Poin s”, Jou nal o Compu e Science 6 (3): pp.
363-368, Science Publica ions.
WEDEL, M. and KAMAKURA, W. (2002), “In oduc ion o he Special Issue on
Ma ke Segmen a ion, In e na ional Jou nal o Resea ch in Ma ke ing, Vol. 19,
Nº 3, pp. 181-183, Else ie Science B.V.
WEDEL, M. and KAMAKURA, W. (2000), Ma ke Segmen a ion: Concep ual and
Me hodological Founda ions, Kluwe Academic Publishe s.
WEDEL, M., KAMAKURA, W. and BÖCKENHOLT, U. (2000), “Ma ke ing da a,
models and decisions”, In e na ional Jou nal o Resea ch in Ma ke ing, Nº 17,
pp. 203-208, Else ie Science B.V.
WU, J., XIONG, H. and CHEN, J. (2009), “Adap ing he Righ Measu es o K-Means
Clus e ing”, KDD’09, Pa is, F ance, ACM.
WU, J., XIONG, H. and CHEN, J. (2009), “Adap ing he Righ Measu es o K-Means
Clus e ing”, KDD’09, Pa is, F ance, ACM.
YATSKIV, I. and GUSAROVA, L. (2005), “The Me hods o Clus e Analysis Resul s
Valida ion”, T anspo and Telecommunica ion, Vol. 6, Nº 1, In e na ional
Con e ence RelS a ’04.

ANNEXES
131
ANNEX A
BIVOLINO
The in o ma ion con ained in his annex is om Bi olino websi e
(h p://www.bi olino.com/).
“Made o measu e shi s”
Bi olino was bo n in 1954 in a sandy a ea whe e he squi el eels a home. Each
Bi olino cus omized shi is emb oide ed wi h he squi el as he symbol o he pe ec
biome ic i ing gua an ee. Today Bi olino suppo s wo p ojec s ocusing on
p e en ing he ed squi el ex inc ion: The Sa e You Logo ini ia i e as well as he Red
Squi el in Sou h Sco land ac ion.
Bi olino B and S o y
A. The Beginning
O e i s 50 yea his o y, Bi olino has buil a epu a ion as a leading supplie o
beau i ully ailo ed shi s – c a ed wi h he ul ima e combina ion o p ecision and
design. Bu ew people know he s o y behind he b and and how i spea headed he
design inno a ion we see in shi manu ac u ing oday.
The Bi olino s o y began in 1954, when b o he s Louis and Jacques By oe
ounded he company wi h a join in es men o 8 million ancs. Thei g and a he ,
Jacques, had been in he linen ade since 1900 and he name Bi olino was chosen o
ep esen bo h he amily name o By oe and linen – he ab ic om which Bi olino
shi s would be made om. They based he company in Hassel , in he hea o Campine
in Belgium.
B. The O igins
Wi h he opening o he By oe b o he s i s p emises, he ounda ions o he
Bi olino b and we e laid – pa ing he way o Bi olino o become a pionee wi hin he
shi ma ke , a symbol ha gua an ees quali y and he pe ec bespoke cu wi h a
ashionable wis .