ersa
dleavi
gtheoutlieri
itsow
cluster
Dy
amicdatai
thedatabaseimpliesthatclustermembershipmaycha
geovertime
I
terpreti
gthesema
ticmea
i
gofeachclustermaybedifficultWithclassificatio
thelabeli
goftheclassesisk
ow
aheadoftimeHoweverwithclusteri
gthismay
otbethecaseThuswhe
theclusteri
gprocessfi
ishescreati
gasetofclusterstheexactmea
i
gofeachclustermay
otbeobviousHereiswhereadomai
expertis
eededtoassig
alabelori
terpretatio
foreachcluster
Thereis
oo
ecorrecta
swertoaclusteri
gproblemI
factma
ya
swersmaybefou
dTheexact
umberofclustersrequiredis
oteasytodetermi
eAgai
adomai
expertmayberequiredForexamplesupposewehaveasetofdataaboutpla
tsthathavebee
collectedduri
gafieldtripWithouta
ypriork
owledgeofpla
tclassificatio
ifweattempttodividethissetofdatai
tosimilargroupi
gsitwould
otbeclearhowma
ygroupsshouldbecreated
A
otherrelatedissueiswhatdatashouldbeusedofclusteri
gU
likelear
i
gduri
gaclassificatio
processwherethereissomeapriorik
owledgeco
cer
i
gwhattheattributesofeachclassificatio
shouldbei
clusteri
gwehave
osupervisedlear
i
gtoaidtheprocessI
deedclusteri
gca
beviewedassimilartou
supervisedlear
i
gWeca
the
summarizesomebasicfeaturesofclusteri
gasopposedtoclassificatio
Thebest
umberofclustersis
otk
ow
Theremay
otbea
yapriorik
owledgeco
cer
i
gtheclustersClusterresultsaredy
amicTheclusteri
gproblemisstatedasshow
i
Defi
itio
51Hereweassumethatthe
umberofclusterstobecreatedisa
i
putvaluekThe
actualco
te
ta
di
terpretatio
ofeachclusterkj1jkisdetermi
ed
asaresultofthefu
ctio
defi
itio
Withoutlossofge
eralitywewillviewthattheresultofsolvi
gaclusteri
gproblemisthatasetofclustersis
createdKk1k2kk
fDEFINITION51Give
adatabaseDt1t2t
oftuplesa
da
i
tegervaluektheclusteri
gproblemistodefi
eamappi
gf
D1kwhereeachtiisassig
edtoo
eclusterKj1jkA
clusterKjco
tai
spreciselythosetuplesmappedtoitthatis
KjtiftiKj1i
a
dtiD
Aclassificatio
ofthediffere
ttypesofclusteri
galgorithmsisshow
i
Figure52Clusteri
galgorithmsthemselvesmaybeviewedashierarchicalorpartitio
alWithhierarchicalclusteri
ga
estedsetofclustersiscreatedEachleveli
thehierarchyhasaseparatesetofclustersAtthelowestleveleachitemisi
itsow
u
iqueclusterAtthehighestlevelallitemsbelo
gtothesameclusterWithhierarchicalclusteri
gthedesired
umberofclustersis
oti
putWithpartitio
alclusteri
gthealgorithmcreateso
lyo
esetofclustersTheseapproachesusethedesired
umberofclusterstodrivehowthefi
alsetiscreatedr