Cluctesing ‘and * Clusters Bpalye® g- ' ee aoe } . TOFD, * Pe apne flas groups of objects together & % Fore Spachistes. 9 | The data ems are clustered based oo poincibles Oh mariner cr) the Folea clace oft dlaaflfes. and Mintmreity hq the foter clase! Stuirlasrlted « B.To ouw clase use dre ‘placthg all the gle at one side and all the boys at one Side Gitle cluster and boys ave boys Cluster - + Gitle ave Tt fs Unsupesvited Machihe Learnitg alqooithm. ; , cluster 4 - © v aa I (| \ eS of Pipe et = clustering Scalabltly ; — Algoo?thms usabrltty uofth roulttple Lypes of dale Dealing wrth Unstructured data - Inter operatab?| Fey.clustertog method's S- = Pastfontng methods | = Hesarch%l Metheds = Denstlfly Based Methode = Gptd based Methods - es Sh le & eloete Analyete : clastertng Analye%s Supports Data Stoucluses, spate otilichetes Grae Of Lod types. They aves 4: Data Matt . a Detar lastly Mab OB" Bata Mabe 2? oe! Here, Data % septa as ‘able (os) fe Pp matsie bev oo! wows —> real world entlia be (names ) Columns —> Properties, of enthties. a Dieefmnflaaly Mabirx ¢ - it & represented ae oxo > ela. i: get beluseen oe objets eolFizc Staten es aay? oe yendtid oh ii) olen 2) ----0 Mypee Of datas vel a hed cag} r * Thee are. Four ieee oh been They ae 4 Thtesval gcalbed dita be a. Bihary Variablesa. Cateqostal Vaotables 4: Mixeal Variables -, 4: Interval sealed clata s- + Tt has continuous varables ExS 10-20, 20-30, --- ete: To Convest thdivPtual data fto Continuous variables. J . = do the tar data anda nate atten before that: * Data standardeation means removal of units + For etandardfeation data we Should calculate _ mean absolute deviation: « Then divide the clata tito fhtervale .. & Bihasy” vatriables 3 ra +h has only 2 dtates | Mf Soe Vertables is > abserst | arab le Ss ® present ° S tas = gublypes Z ~ Symmetie bilary states of Yaotables can RSs, change ribet A? States of vanfables ies fngantt Change pos Categorital Varfables ¢- Data that can be, dfyfded Sal Pe ~8 a a Toto cateqories eohel af,ra 5 nominal vaviable 3° tt, bas no partteular Ordex to the Meee: Codegprtes ‘Exe Geodeo / \ male Fernalé (can be. Fo. any Order). b) Ordinal vartble ze TH hae pasticulas Poternal jovdes to the Cabegosie Ex? Temperabure oe ae | tow Meditien High : (should be 5 oxder) 4 Mied Variables 2- Combhhation of Afferent types of variables: Cateqorimation os Mafbz clusteotng Methods #- Te cluctesfng methods axe categorized “to Four types. They aves L Pant Fonfng | Method 2: Hitzavchial Method 3: Densfly- based Methed 41 Gpid- based Method. £ Rationing Methods go.) + The data © awitted, tate parthleos- Fast? tfon segs pent a cluster: Bechaclobtes should be less than ©) eqytal to total data Phe”2 Past®bfon Should past@fante satfefy @ Rules. Leach pasties should bave atleast one object. ~ each obfect should be belongs te only One parity, Exampleg Mean algartthm- ~ data % A®ited Blo clustedve based on Afstance and centro Values: Hight (x) | web oo: ) 125 tapes OTS 2) 1t0 fe Bb 3) 16@ ga 4) 144 ce > viéde the data tht two cheeters. Each Cluster. should have centro? . Clustes | s a en cluster a : (tes 42) . (190,56 We Can take any Order pate From cxample tornlow Pemathing values should be Aw Red based on Ecucledtan distance (tp) ED= | (xy-%e)*+(Yo-Ve)™ Rs) Kis \[(ies-tes)* + (60-32) = 208 : : V Rie a (tee (a0)" + (60-se)* Kirt ky " 448 . Tt belongs 40 clustes a- nlow, calculate the new CentsoPd) | (140,56) and’ (16,60) I Uto +16 ( 416g » eter) = (14 58") z a= (169,08) i a sila Win @ | mr ®, kis (osa=ea say = Gar | ' (it4y6@)a15.- wae \hl4—leay” tGe-se)e = last Bocsien meek art I + Tt belongs to cluster 2. calculate centeotal fos’ (ies ,42) and (194,60) i | fiestiag 7 as { (eset as) = (e2,40) The data “Ei arte Moto two different Clusters.2: Hideeachial Methods & + Tt quoupe the data foto tvee of Sat + The chuckuse called. dend sograr: . Dendoograrm have Sequences ol merges and, opl?ts - © Urrarch&l method bas two sub methods 4 me were Method DAP WeMethod- 4 on Method @- + Tt % bottorn —up methods. - Steps Involved 2, - stort) 4) caleulate the eft tasty ae one eee earth veepect 4o all other clusters. a) Consider every cata pofot a Tod dual data. 3) Meage the clueless woth he hee! sfinflasfty: 4) Recaltulate etenPlavthy Fen eh, cluster. 5) Repeat ® and © untfl -the Shogle, cluster % obtathed.ee , These ave & modes = Sfogle: lmkage. — nfo < —Complete lfokage — max - Pverage likage ~ AVF- a: DIVis%e Method 2 + Tt % top-bottom methsd. «we take all the clata Pleme foto siagle cluster and f Pleratfons, we Split the data- «+ At the end, we get Ni clusters 3+ Densfty- Based Methods 2- + Data obfects ave clustered Bnet on dlenstty- + Denerty means mass /Volume! Example 3 pescan (Densfty Based spatial clustesiing of; Pppleations with Noite). ve tt bas a frpute Geoatad) >é - Epelen seifhiinurm tpats polite -Lthese , ee E — radius of cBele aah formeck whth lata obfect> das centee: mfp pothts — mfrfanun 99 O data potnts thee the c&cle Exe min pte = 2 Types of data points 4: Core pofot 2 ; Ik Should satiefy the cond?tfon, ph min poldte. 2 Boundary pot : nefbbour! & the Core pofot: 5 3» Nofee point 2 It % net cove nox wef: boundary pofhts - Eee .4+ Grid- Based Methods & yp Tk usecia mull? recolubton of qx data stouckuise . web afiated the Obfedte. ‘Foko ‘Forté o6- % celle that form a grid Ihe Steuckuse. + - ~ then denetly fe, calculated fox these celle aieort the elle acon bo denefly. — Tdentity cluctes centers Sosop Update nemohboles ‘celle. ' + GoPd Based aoe) fe Qufck proteeethy tno Example g STING (stabttveal Information Gal clasteotha -Algorthm) ~ Spatial data AvMed Flo vectanqular cells ot Afferent levele of aesclution, these celle foom a tree steucture- gees lseg - cells at hrghes levels contafos Small celle compasel to lowes levels. - Clustering done paced a 0m pasameters ‘mBan }2ount min ymax. = Caleulatibrd of these pavametess should Stast feom oot and go down til bottom layer.Outles Analyse” —S , there! m + Armong ther data, Peme % Data Base eo ay ty fame, which ofo, net, Fellows the qentval bebavi, Some Per 2 , -ur of datas + Those data fleme a6, called antlers | nown outie fralyet, . Palyate a “cutlen % Le Outlies Detection & 9 a The procese of Ment?Fyinig outlitne and: mubsequerty vemoving them. _ } These se tus methods, They ave? to statBital Approach teal, ProPitly Approach Le ctalfotital ae bskivi eles ro » ett! &'Based’d0 poobabslt Pty data pote. . Lecoett Auge ite: iz conctlened as outlite j <"pasamels% method - Non Pecpnetate swethod &: Proxfinflfty, “waabs ttt fe based. on locaton of data: pownts: BindPly bosed Approach ~ Dfetance based Approach - Grd based Approach ~ Devialton basect Appsoach.Types Of Outliers 3 These ase thee Aypes of outls, They ave: 1 Global /potrt outles # CollecdRe outlven 3 Cond?tidnal outles. 4: Global / pore Outles ¢ wlhen a stale data objects devites oom the vest a ) F data: Pointe Global /pot oud en. acollect®e outles ¢ then a qroup of data obfecte deviates Dom the © vest of data, 2 cond?tfonal outlér¢ Data obfects deviates from others because of the SpecPhic condition: