Appunti Machine Learning

AREASdiloep t t uboncol loT moke tpamRESIDUAL SCRSSoh)RSTA ACTUAL UALOERALNIGSE7HLeeREL1cteullkolid uAUEKAGERssWokuaSDOWE ad uohbepiu o eoupnahioC o aje XQud edcukate RSS mokiu onRSS PLOupaOtla opliaro Ico b 1eu ondtl SOBTLED2iouo eeOneT LEAF No Cacodimue plnaupuctl UER HEH VARIANcE na Bedoua Ihle oTL o clom eNt doo outwt No ERROok,dlurly VEE ITUNE STTUATODNDukei w ekcta uddusa CoTbu poiat meedealep oddupCadnd nodlio khe la minmloe modeslo oncud20U uL TLE TEATURESI ek on val PetodpLoit oCoue H SSRwtk u TUNIONVGCalCuASSIICATiON TREENou Cdopee d lappk e d to buld o wpedolumteoio punlponLe CUuo4ole oAUoTa o tole)tiou ate CoNgbeelhet Ha bonuB et ue md mbt ouneret t uuel louea oi) dushbdtt oueuado)T meo otuheu e o k aolD PLoBALoSunpiune t e a ENTROPL ootmeou tSORPRISE uwoutu oeketLot ebpo ut o peokob.tT tioaedaopyceled.uexpeeted aunplue beLe cupd UFORATio CAIN edel otuateoua tstole bep lnuody elomteol ouudtta 6uh 2eundkauTLocdt tu Jola joid, ove ruodlg bee elonied doudCodthua udue loe d

dolojoude be LeafudiotA 44-1-2024Hada CuenaLioL 4-23-44-2024 TREESWCASIFVALUEANDNUERICALCATEGORICALLost tiu we vRKdeutlonCloricocuudpuna Troeelondt.otuueataduHu eonat tuop eta-Tod.ou e wour tol tol mo oo elonr cotighoaoud uboiicloWhat halbeu wuuouhoue t wic@ od leeg atuuo do oveS apdu LABEUS b uoa houbewtits oubire Ue ulewbleo aolund(H pdalug )T yon loveCou-pde ta AERAGE VALUE olroo 9o Coucokegoraze avolts ooaie oCtou VolinOud Uoluas benoudoTLoLr Hla PeotunKPARC AEO GAIN dutt e eud NoLeeo eleApa2,5 mre llhi d puedud 12 5 Mmoole eh 7 fe c12,5Col cololoPusSTNG ALUESWe couhuie tustiwhiza oru Hm ValuaWISSING CATECoRICAL VALUESe o uep ADD THE HOSTFREQUENTI UALUECa d aCaIGTeG5ueaud une T pLa on eFC Chont fouu0pluidilua to BaikedAstei")|HSSIUGONTUNUOUS VALUESSa eo uplput e UERAGE ALUEou eo d 2Aeeyeosup4 buld KEGRERSlON HODEd ouophediot t l mnnu VoluPRUNINC RECRESSON TREES mmooldlaOVDtalcdcoutwaoudTa quato HouTo BUUDt ut a o o pee oleyHCracyhapoeQudneotetSSRHacoupdeSou

omodermbumoldeo2datohoveUeLoveto u RAE GF betwve tlwSRoud a GENERALIEGENERAULEATIaU POJER Thou to copke TRE cokilEXIENALTTREE ScoRESSR T Coueaty TeuslySSR+ o-befac cdoUi la ( coin) meosoUILL opaod otoe betwe etcoud Peunolaot.ou bdwe o a thHoW e We eoldte Sued ote eoumduue data aaholi RAIN O TEST et uo tuoo ALL DATA)oud eAbluoke o duou TuopulmeUle bulGtu tae oud Vanodasunedl TreeTeST uoutaedDdoSETTESTtlucouaiduetoaueukea [olieE5ttetrecaidSRoudtuCopudlehele-toSSR d numhephiztleoluitleoehaiewwestepotWesedAtKANDOH OREST tbbididandeoHe,oHtow iuetdoocoubita ceouktT o To de l 2o coblo EcozszsA PeDDATASETSE dätoer lauiupta mumbee topiaHaoKddepet oi9 taa kole tla oatuethol oneBANDH SELECTED LD 2petitahoHla_oipiueloTHe Hvzs bu.keld bu CoLdiekePeskides otom t by couaidiaupSOBSETS otuAo (exCooiduno Gol Beoad fauesut oud Bloekcol AutuTo buSOBSET OF fEATURE SAT EACH STEfF dlem Ipo up e I e buloExt 5 o a 6 tiws hole aCetoau o u po tuo Aoupe uutl be elome TLA WoLTw cdle BAEGING eveunugk tionieR

olet edcicunohr luodlo buuCtoe boolalhobed dooeelBut HocaWe oludle He eCURACT) Kouda foTLadtaiCt OUT OF-BAG daleeaeb dld Hla edeS EA TREE V udle Hu OUT-oF-BAG doorels oud Goptl MSCLARSIFICATIEN ERRORa t hQd-o o doteeVoni olcei HMPERIARAME TER= du fheo oloin aCrekat.ou ou akadl AiteuloocuL12-25-41-2024Hauds ouJepoou 4HDeeloo a eudou13 4ola 202150-44-Hox TOuILD A HU SYSTeFIEd o Cucro Too.c i KL iou eePocuged otTWe ok darect bol. REARE SENATION ideudptu HYATHESIS SPAcOPTIMLATUN ehaneENALUATUONE cheomou oludt.cul o duptiupuea aruGoAD wwank i Cdea mod LotuokeaoodgeunolzdoPabu eeeu exoupeo (thou cetexoupkwoaceaduatuie oKL Asne -CadlDATA ANAWSISod PRE-PtoCESSNG uatol DstodLoDEL LEARNING Test tEVALUATLO New DTaAffuicATTONDATA ANALkSLS R-PRoCESSINGoNCOHPLEEE miu olukAINACCURATE wrop uoluwzno-Col GIGO4pholeo CuEAN euz oloto Re-PeoCE SSINE HoeDATA CLEANIG Ddtelors KUuciueNotse ReuweliugDupicoles uugDactceCHANGINGDATAA NoAto oud KEVdlugCKEATUNG NaW ATTRIBUTES (b a i dpa d eOt&e Reueio exop

Box ptat) Cou 9uoatllhColod OUARTULE S(a pttolusd dotoedo omoh oCodlho laue ECIEN od GERCENTILESDExopk o aagokeeo copt uateu 2o5TRF1,5NTERQUARTE tluKANGEduaeldToovegL QR25-1254Eod 4betuwenduabd-T QR wped tdouduoUteo dola ao mVolue 1,5 45 4 445 po 2 ia c oleeRacagou2otUo2moliHIN-MAX NOKHALURAT ONX=X-iti 6 - ) +amox+miu2SCoRE NORHALZATOAXEXmRon(doviol(KTeckueek.cticuHowhe t d eotunes ue hote ora uefu eea2eualeOr Ao oue otenWe eo ue dred toolsPoopl EDuo ExpeatRten tuy mioAu tle upoltone eoehpoune eo. LmomdeeoiuETa Tuok oonoo)wwalolty edetio tmulo to eota ncolunp ad aosud 2ottia oAuokun ESVD SugGLueo DecopontioutVALUATUNG OUR HPOT HESIS RAINNEbuld18tieodataaBe/20o4pKW wouall onSETtuolnTESToudzmcoleltonouSET e)ERRokTRANTNGHIE2Eponouatenotat ueuoeobeloae (a Tnoudepoewe haue am enoma to0 tutoinetue "te oVERPTTINtu dotahaje a NDERESTIHATLON o t euenoltotiou ennca lo alkou ou a moolk TkAINING SETIHo ed SETUN SEEN DAJ ddeAmud HlDHbo TKAJNING SET FCao hodil20 CoS-VALIDATGN SET oogpe ED

HEL(at ne dtieo EST SET TAcua lbeill madelACROSS VALIDATIONThun o d CTLcda To buac e VoidlotoetHoLD-OUT CtoSS VALIDAT ON io bucet uaiu ( TRAIING SETK-FoLDS CRTT VALIDATLoN t wiCe be en eHoLDoUTOudtat exo uou ou ote e b UNRALANCEDDATASOT uTeonou IVaree molla lod u tlostoepuKtoLD CROSS-VAL IDATUONpiwol dDoi d e d (eguele auboln FoOs)2apedledKtuh Kwult ho tu Ado Cobe tlan OUageolto 2oducrupk imatiou Uwlenk-1ohalugiziage dbraok bus o uoel (a thectg.cdest audeoeh doARAVatL Ued teXocteyOmRANDSuRSAKPLINC looudhodeedtodolapoitu.odaudcuue Xpensmeudbodo od Co pedlcone H MUulbeHou TO CHoOSE THE AEXT H1ROTHESSA HL DIKGNOST C uo Co Zunpo t wdAul eXo dootie BAS/UARIANCE GRAPHPaulictonaomees HLcRE Coupex c d CosOVENE TTuG hode a Hoo cokpeak tmale)ha dspluo kc at vol loLATRDA kpdou totiTot CURVESLEARANINGouolanu o timaluowzwmadlekoredeodTodt_ m 4 dateoid o-aVoldlotionlno_mmcolelueoulthonuigaCoupL t eo)JThomtodCal inue datlo o( tuEARNWGCURE o odleR Ne0be HICHBIASSTARLE CowDITIDA) Cuieo Cotinertoeg oc

o bold up ooe0 noupes w hutouetathoidtp Cuie (SHh ouWot (NoTALNAADDWG NEW TAUING DATA ihep ueGH VARTANCE Ouptpuon ooonpo p0pPetuwe Aocded d olddliaCunta HG*CeVARIAAS 0FiTIwGUoldoftootmtnmpUol.dlup AD C do oa lugo6LCuae CoiupnenebaeAHu yoeo ADD NEW) 20upe auuel aibuKENUGVARIANCEpetue Gdauticd Veatu nblaecouphtyolVoJupula14T02-12OrG&A RTudagsuaturHO LOHEASURE HA HU SIS WoRKS UUELLJe dtoeto euoluoetu tolmaulcda HAEKowmo RHSEAeouRaCucnoRcolfoc LeouceSSiONERRORIEGREeACHG 2ca dudNweTtooudtVoluesued)etsweaxin HAEHAE Abrolde Eme ho a ouplbsmiSE MonSaLuoud Ese HSE 31RHSERt fuu Souat RtSECECisoN BoUNDAND od coul tu mumber nuaClomed) dore@w lecdto eoushueo ohx tl ho.1h o ho pocd) uolomiPorla CoNFUSioN ATRXEACTUALCLASS eSs, t tCou Tksepok ve ueetoatwe TH FEFN comede a bunPREDICE fortiueTCLASS NdtioFN TNAccuRAc b Corructeg elomied oualpPTNTPTN Fr-FN ollnoueaPLECSONE -Dmo. comed mdoted oupleTP+ FP o potue hedkdeoRECALL SENSITIVIT1 valisprecletedcomuëteTe no.TPFN le Fotive nounpeFOSITIVE RAE

TNS (True Negative Specificity) = True Negative / (True Negative + False Positive)

TN Rate (True Negative Rate) = True Negative / (True Negative + False Positive)

Error Rate = 1 - Accuracy

Accuracy = (True Positive + True Negative) / (True Positive + False Positive + True Negative + False Negative)

Precision = True Positive / (True Positive + False Positive)

Recall (Sensitivity) = True Positive / (True Positive + False Negative)

F1 Score = 2 * (Precision * Recall) / (Precision + Recall)

Specificity = True Negative / (True Negative + False Positive)

False Positive Rate = False Positive / (False Positive + True Negative)

False Negative Rate = False Negative / (False Negative + True Positive)

Positive Predictive Value = True Positive / (True Positive + False Positive)

Negative Predictive Value = True Negative / (True Negative + False Negative)

False Discovery Rate = False Positive / (False Positive + True Positive)

Operating Characteristics:

True Positive Rate (Sensitivity) = True Positive / (True Positive + False Negative)

False Positive Rate = False Positive / (False Positive + True Negative)

Positive Predictive Value = True Positive / (True Positive + False Positive)

Negative Predictive Value = True Negative / (True Negative + False Negative)

Two FICANT SIGNAREKESULTSHEIEHEASUREHau To tun l t o atethotauowemuehoRouhLeewto boteP a c t l a t l o u l e u AttoduoiuedstoueWemonea 6 we hove|ekoletudurotoudautluouol.ted u A oudTulo ta hobl SiGNIFICANCE OffRESOLTSlo aWe Tla ncb alotioticiousoe 2eoaeo2 teekuaass bbwtu uurtlot A otE) tlal BCouARIIGSus USIANG A PAIRED T TESTHUced Tto datenmeTlu otuatieciouelae deu 2hpotLai-NULL HIOTHESIS.= 24aruinno pue hode tu SAHE ACCURACYAUTERNATIVEH4P=aayo n me oceuroleto tu ctere Coutiamtu UOUD HHE s mioup H dRdosevolee dluasotexpCorneo b oudo

aJUOA lOte usl ( Ue caurto40,05) th NOLL HYP oadlAuout duteaoleate bitayoyoA ad8a t opTeemalo oe AMCLE HEAN2CLoteHESTATISTIOtatokstie petockdLotve toded t Accuracu tha 2 Autetauou Cageioleoh tdu to clouc Couko addlRDETERHINATIONOCoEFFICIENT tlahoveLenicumalilteCweajohtt aaun olaiiuoktlexTeuncuaome gomououdUoruupoptuodotomu@kou(holuTtoit io exblalned bute wou pLAL ol1ot