Dt- 04/0 a/
Boblems o k- Mens methed -
The k- means alg0hhy is senstdeve to outliens.
9nee an ob podt uic4h an amtHOWely lange value may
Sub stan t'ally di'stt 4he disteebntton o the oata.
- k- Medds:Inetead o teakingthe mean vad up o te.
otbjeet cn a eluste as a ne erenee pont, me dodds
Can be u od , whu'eh ci tthe mest centally
loented ebject en on cluste
Dansity Baued clustenong Metheds:
baed an densiyoeal oustox cniten'on),
Sneh as -cou neeted oints
Majon Yealunec
- DiscveH elutens o en
- Handle ne'so
btrwy shape
Ceno'isn
Sevete) tntenesteng stud'ee
DBSCA N
" 0PIIe9
Denity coneept :
objeet Ceo): objee! wth atleat M'abqecs wilhin a rad'y
. Dreely dens eaehable (DDR) : x 's o, Y en xs-hehbeakd
Densiy seathable : there emeste a ehacn a DDR objele
Density based elste i den ity cenneted oyects mamM
[Link] echabi lty
Densty Base (( DB) Sean elustentng Method :
DB stan alqoihm it equlree a paramekens .
Pseint ween bwe
means ¿h the distanee bo
ps then
Paint ci loue oH eyal to Eps then thy
boarde pocn
0-out
Core poht/
-1! the tps value ts choosen vysmall then lange part
oda ta censtden as Bntliers.
1 t es chrosenvy lrge then the cu ster will mehye
and ma0hy eb tapoint wcl be eh Same elus, bea.
(2) Minpetnt / Mon pts : Men'mum no-a peins/ Cdatapaints')
ne'ghsow
uith in the tps adous:
the dntset, the baneye va lue ot minpeint must
be hrosens 4
,the mu'ne'mun minpeunlCMIN ps) cun
Aea general Hule,the
Aved nem the no. ot e'menst'on 'd'en he dataets
ot di'mnste
Hentuhe)
as in pein qeatu than or enal
1he miilmum value ot mlnpint at teat 3.
Bonlepent: A peint wheh es ewo th an minm podnts
w4hc +he tps but ch 4he' n'gt bouhnd
o corep nt call e bo wroer pe-tnt:
boer point .
Steps o db sean algaithwm
alg
Step-1: Stat wih an tbilany stanting' Po'nt that
has notbeen wsited.
tmtaet/etnve the nevghbounhood oths
&p-a :
Cal pa'n ls hteh ne wthin 4he e p&t lon d'stanee
sushieent necghbeun hood ano und
e &lat ano tae
the pe'nt then elusten'ng pocess s leveled as
Pa'nt 's mank ed as v)ted else
neise a lote this pent beco me part ob eluster.
step- 4: 1t a pednt l's ound to be part o thé cluster
then ts - netg hbownhood alio an o7 eluster &
the above phocedure Hom step-a is epeated bo
all the e-meeghbout hood [Link] is Tepeated
ntal all the pe'nt s wa'ted.
step-5
leal dng' to the ywnther alutenng /ete e.
1h's procee conhenues wnel all the petnt ae
manked as vi'tted.
butweHg u thoseata ponts that aee
CignieenHyideant fHom the est o te toatapait
nataset hey ane oten,abnoHmal obsenvaton
4hat sew
stRe the data dttebuteon, and aige
due to nconsetent d a enttey ot eeoneous
’ they caan
Outlie deteetion - Date-at/63/a2
eot ef duta undet anal
- In stat'stic, n outtit s cn obs2batian pontE ohth a,
distandt fo other ob sevatis
- lns ttu me ntl eeHOH 0H moment e D h
- Humun mi'stakee o eneO x. CDota entry)
Syetem ults.
Empeion tol eSH CData omtreaction, en peru'ment plonnine
- Date procosstg enHoH CData mani'pulaton)
Sampling' enno Comtrueling' on wieag
m'm'p data rom oTeny
Tg
- Mis repercting or nlenional
Untvaient :: n be found tn
in stngle eatune lpare .
Wulti aHient : Can be tound i the n-dimentlonal enthe
Deteeo o utles:
3teehndques
stalisti'eol ArpHouth
Dtance besed appoa ch
Dervaisn based appebach
StatsH'cl Based ute deteet' on
st wsuwmes a distibnt'on on a probabilily model
given dntaset cnd then identt}es the eutleere wa.} the
model t'ng a di's camdaney test.
hyp thesit lotng
