You are on page 1of 13
ie plarn hecw Dornetple Cormponerst Co % ued for d Eesrvallhy reductton « >” TY Prnctpal Conyonent analyte , OF PCA % a demenstonallt reduction reethed thal % oflen uéed fo reduce the d tiengtonat of faige cata gets, by banshint fo 2, get of “vaitables mio a Smaller one that eifl! contains poet of the snfomaton fr the large Geb « Padratpat Component fynalis’s, f¢ an ungup erueed arm atgotthm that 96 used for the dimensionattiy reductton % machine learnt OG a Bratistfcal paces’ thor converts 4be ervatfong oe conelated features nto @ set & early uncorrelated features hese Netw fanefomed -patues ae called the pornetpa) componen!s . Folurtion + letra ceuelop an tntutltue understanding o} pcp » Guppoee You vote to dletinguteh belaveers different food Gtemg on dhetr nubtifonal content * yp Wibich vatable uAll be a good chote to differentiate food Geng? to can Create OD autffPca) aria ble thucugh a \enear ou aiisn of ofiginal vartabler It¥e Beery = YdVard - Ga ones 45% Vara > Suppoes ke have, to trangrm a a> dimen tonal repesenaten . sco pink to a ome diroenfonal representation § 9 ut ib tnd a ecrot nh Ine * e mn sttanéc 4 - seth a aes the Eig Value Cecom poet Gm Verear Algebra , eigen lecompostiton Pe the factorization of a Mabie into a conomteat form, kthere by the mabit a repsesented tn terms of its elgen values and efger vectors, Om tN (1-1) (oni! -2 -3/ [a zJlo4|(e- ot oigost ta, le pros mantt mabe abit matty 4g: Etgen Ualue decompoetifon => common terms used 7m PCA ¥ Dimenstonaltty ¥ conelatfon se oxthogonad k Efoen vectors $ covartante Matta Realfons of parfncPpal component Analyse PCR 96 mcinly used ag the dimenstonaltty reducttor, ue I vay fous AR appltcatfons Such as computes «image compression et ako be used for finding htdden patterns » pohere PCA #8 used ae Finance, dalamtoing , me @ Linear Oiseviminant — — oe # Linea" Ciscrimntnan a earate ‘ is one of the commeonty Used — dimensionality seduction’ Lechmiques in ml bo Bowe Weneventiaam itcso - class ctassifiéoton Problems. 2 Te fs also ienoesn as, Nowoal Discximinant AMAtys Ss CNOA) oy iseviminen.t Function Analysis (OFA). # This Coq be Used te Project the feotures of higher dimensional space BIS s Sedo inte , lower dimensionat SPace, bel ht 4 ‘ Bisraieno- ewe oy: 40960 classes and (ole teed thera efRicientty: classes’ can pie featuves- osing enty a ture te classify ther may overlapping as sheso betoco: ese coin Keep on the no-of fea tives Pow 9 Howe Linear Discriminant CS eS ee Looks 9 Qo as oO binenaiogall wgdasteg ne technique §9 machine teornirg- we com easily transloca a 2-0 end zh oo 3-2 graph inte a td ptone. Let's Consider an example chere we have €us0 Glasses |in a. 2-D Pane having on oy ans, ond ee need to ctossify, them efhicterty. AS 8 Vote atready “seen a the above exampie Bae ton “erobies Us to drow a \ i that. completely SePorate peers LGA Uses on x.y fate. new. axis by | Hehem using..a Stoight Une wwe Con reduce the To create new ants, LOA uses t+ Te maxtmiaes the distance b/e | means of tue classes. 2 PTE minimizes the Variance within the trelividuat alass- sing these tuse conditions, LDA generates a Neus aws- Unies : © tivevarchicat ctastering e a tHerorchical clustering stots by tenting cath data points a6 on iecdavrdua ctuster- . ao Heyarchicl clustering } clustering js an ofterttithes appwach t ikzmeans ctusteing for i gues ia the database oh “there ove 2 types of tierarchicat ctusterng t 4 te Aggtomenative tierarchicat ctaster: ng 2. Divisive clustering & Aggtomnenative nierarchical clustering Aggioneatve ctustering is one of the most common types of hierarchical clustering Used to group SIM tar objects: » Agiomerative clustering is aise (nace ” Aigexith 1 Detereaine \ gre. similarity feo Feabviduats | clusters: and at othe : ata point as an fp. consider cael cach pace cluster 3. combine seniter cluscer® Recarcutate Proms otuster 5. pepeat SP 2 Bh Ure You get fo singe cluster: on matin for each a. Divisive Hierarchical clustering —_ —_—— oDwisive Merarchicat clustering is exocttes opposite of aggtomerative tiverarchi cal clustering. e Tn Orvisive alu tierarchicat Clustering au data points ave considered an individeat cluster: » I every ftevation, the data Points are not Simitar and separated fom the clerster- 4 Eimaiy we ove tee with © Clusters. © CO _ eran) <— Ey) © steps step, SeP3 —-SEEPR Sey Adwontages a gimp to impiement *« gues best output in Some cases. Disadvantages a” Tt brea the tage clusteys. a OFfficart to handte different Saed clustes. Ke means Algorithm wes means aigorithm, where each cluster is neprésented by the. meen ‘vo lKe of. the + objects inthe; cluster. | 0s % K-means Clustering (Ss the mest Common + Partitioning atgorithm. ¥ The’ Kk: means: atgovitm. is implem ented «in U- setps: +. t Pavtidion® objects inte Kk non ~ subsets: ? emery 2B. Compute feed poidis an the comtwids of the ctunters of the current Pant The centwid is the center ot the ctuster + B+ Assign each abject te the duster With the neayet seed point 4 Go bagle to StrPE, ctoP when No move nes antigament: 2s. [2-m,|* ; "aces “The betes diaguien’ explains, the ‘worcing of the temeans clustering algorithm. Belore temams —") Afier K-mans centboid. | \ See 8 ‘) Outlier Detection - Hit omtalll Cut lier » ' : : Guth “Data potritr that ‘are spnifecanily diferent from the seat ¢f the cata set + fs catled Cuttier. Nn anskaia oh Gutliey detection also known as ener anomaly detection . , tH tsa technique eded én ML ty edentt, the data points - ’ : id : Cutltsy, Ue HY, unuswal evtth anrajostty GF the data \ ; 1) Statistfeat Methods: Dz Kove : dz oO ls anethoct rneallewie ‘the Yetaten _- | 4 a data point from the mean % team Of - Standasid deviation ' | td) Hodified Z- Scowe:. \ ! Ye ts Stmtlan to Z-\seowe method, | but ? ages the median and median ahoolule | deviation (MA 0) | tii) peccentile: At clefinee 2 threshold & the | data déstre bution . & Distance - hated Methods: O” Gucledeam PDbrtamce: alt Computes the déstance between a tata point and tls meanest retobous. tt) Mahalamobis Distance: He Canscelation beteveen vaniablea and TMeasunes the dtstance a a data point fom the Centroid af ‘the data déstibullon: , > Dent - bared Methods: ) Local Outtics factor dor) Ge the denstly devielion a a data point Compasced do tts etgh borg t) DBSCAN : Aentity - Based \: Spatial chistes | @f Applications cufth Molcae - ? Xt fdenitifies dense segions tm the alata and classéffes porati - fn, spasiae segions.. 4 cluétesing- Based Methods: a ) kK Means cluateniing: : After eluatesrg the ata tnto qroups,. points that eo not belong to omy cluster 9) uéeuauchical aes +t can be tdentcfied by examintng the heéght af the denduogram, D-H Bated-peethads: ')troution forest: Tht Qutlter detection amnethod depend ° mratiuue of the data. Non Rewmelric ottgaeston ca — : “ty > Fb fo a machine learning, technique, turd fo modal “he Aelattonship between variates” withoet making oe) ASSUMPHONS about the functional By of relatlo, > nor paramilric agaesien toes not reqs’ piffteg ai Fixed See of Prramelers > Tarp nenparanetaic veqaesston , the meet learns the melattonship between the Ie variables (Lealives) hel thu eeeput yauatle (tanger) dinectty from the dara PF makes rimemal assumptions akeak the andaliying data diststoution and capture complex nonlinear alate . patiens aud ¥ several ular nenpasarrdaic segserston algori thy a pe ¢ eons ased Tn mt cee 4 Keane] Reqvession 2s Ke peamest Ketgbieaes CNM) 3-Detletn Trees & Bupeoat vector Reguenton cevey > Gawslan Processes © bernel Regression : Te esttrntix the tooger variate basct on evetghted ate ote oer ghoouw data potnts, = _ higher woetghe to nearoy potate dung lowes wo. es : a fo distant potnts, a: Tt fo wes to dots, Mine the wetght assigned to cath date. pot q q oe — Hewearest welghloours Ck 1) Zt te at laovtthm > ft te simple algoot : : > Ft preach the target voxiade by firdhing the K clogest training ZaMPl& to the feature APace > the prcltcted vatue fe average on mayonily vote tothe De are t hefghbours Oecigion Trees + > 7hie usd oy poihonh sy the Fectune pace foto smaller pegions. avol asaigntra contant vale te each region “ Buppoat vecton Reqaesion CEYR) t- Pit fe wed to Seqessten pacttems > Te aims bt Bing Pari fn ePound the wypeylane that mancimigce the Precttcreh alugs, wctale allowing a Cataio fevd ox exsos Gausian Paocesses i > Thee are Mexthle nonparametric searversion Me thoof that models the aelationchip between nardakles cme QO déetoution oven fanctons. PF Capture Complex paltins ¥ tence te under nonlf recs, These methods ane poteutoaty weful etn try Meleaicevhe Bon vaxiatles On comple x 3 OF Unknown,

You might also like