fast condit densiti estim quantit structur dactiv relationship method quantit structur activ relationship qsar deliv point estim quantifi uncertainti inher predict quantifi uncertaini qsar predict predict condit densiti activ structur point estim condit densiti estim easi deriv predict interv activ paper experiment evalu compar method condit densiti estim suitabl qsar model contrast tradit method condit densiti estim base gener machin learn scheme specif class probabl estim experi show kernel estim base class probabl estim random forest classifi highli competit gaussian process regress take fraction time train gener machin learn base method condit densiti estim good fast option quantifi uncertainti qsar model sentiment knowledg discoveri twitter stream data abstract micro blog challeng sourc inform data mine techniqu twitter micro blog servic built discov happen moment time world twitter messag short gener constantli suit knowledg discoveri data stream mine briefli discuss challeng twitter data stream pose focus classif problem stream opinion mine sentiment analysi deal stream unbalanc class propos slide window kappa statist evalu time chang data stream statist perform studi twitter data learn algorithm data stream fast perceptron decis tree learn evolv data stream speed boost divers densiti learn abstract multi instanc learn bag instanc singl featur vector paper revisit idea perform multi instanc classif base point scale concept search point instanc space highest divers densiti computation expens process describ heurist design improv runtim result show simpl variant exist algorithm find divers densiti maxima effici show signific increas accuraci obtain appli boost algorithm modifi version divers densiti algorithm weak learner fast perceptron decis tree learn evolv data stream weka da machin learn workbench data mine studi hierarch flat classif protein abstract automat classif protein machin learn import problem receiv signific attent literatur featur problem expert defin hierarchi protein class exist potenti exploit improv classif perform articl investig empir case hierarchi compar multi class classif techniqu exploit inform class hierarchi logist regress decis tree bag decis tree support vector machin underli base learner compar hierarch flat variant ensembl nest dichotomi shown deliv strong classif perform multi class set present experiment result synthet fold recognit enzym classif remot homolog detect data result show exploit class hierarchi improv perform synthet data case protein classif problem base recommend strong flat multi class method baselin establish benefit exploit class hierarchi area index term protein classif hierarch classif multi class classif review multi dinstanc learn assumpt multi instanc mi learn variant induct machin learn learn bag instanc singl featur vector term commonli refer supervis set bag label type represent natur fit number real world learn scenario includ drug activ predict imag classif multi instanc learn algorithm propos mi learn method relat instanc bag level class label type relationship instanc class label earli work mi learn assum specif mi concept class drug activ predict domain standard mi assumpt guarante hold domain recent work mi learn concentr relax view mi problem standard mi assumpt drop altern assumpt consid state assumpt relat assumpt propos paper aim clarifi altern mi assumpt review work area larg dscale attribut select wrapper abstract scheme specif attribut select wrapper variant forward select popular attribut select techniqu classif yield good result run risk overfit extent search extens intern cross valid althoughwrapperevalu tendtoachievesuperior accuraci compar filter face high comput cost problem overfit high runtim occur high dimension dataset microarrai data investig linear forward select techniqu reduc number attribut expans forward select step experi demonstr approach faster find smaller subset increas accuraci compar standard forward select investig variant appli explicit subset size determin forward select combat overfit search forc stop precomput optim subset size show techniqu reduc subset size maintain compar accuraci condit densiti estim class probabl estim abstract regress scheme deliv point estim essenti quantifi uncertainti inher predict condit densiti estim predict interv deriv paper compar techniqu comput condit densiti estim class probabl estim estim appli discret target variabl deriv instanc weight underli univari densiti estim yield condit densiti estim densiti estim compar histogram estim previous context normal densiti estim kernel estim experi deliv perform term cross valid log likelihood term qualiti result predict interv empir coverag interv close desir confid level case includ result point estim comparison gaussian process regress nonparametr quantil estim larg dscale attribut select wrapper abstract scheme specif attribut select wrapper variant forward select popular attribut select techniqu classif yield good result run risk overfit extent search extens intern cross valid althoughwrapperevalu tendtoachievesuperior accuraci compar filter face high comput cost problem overfit high runtim occur high dimension dataset microarrai data investig linear forward select techniqu reduc number attribut expans forward select step experi demonstr approach faster find smaller subset increas accuraci compar standard forward select investig variant appli explicit subset size determin forward select combat overfit search forc stop precomput optim subset size show techniqu reduc subset size maintain compar accuraci human dcompetit tag automat keyphras extract paper connect research area automat tag web statist keyphras extract analyz qualiti tag collabor creat folksonomi tradit evalu techniqu demonstr document tag automat state art keyphras extract algorithm improv perform domain algorithm maui util semant inform extract wikipedia maui outperform exist approach extract tag competit assign perform human tagger analys chromatograph data data mine monitor petroleum content water chromatographi import analyt techniqu widespread environment applic typic applic monitor water sampl determin petroleum test mandat countri enabl environment agenc determin tank store petrol leak local water system chromatograph techniqu typic ga liquid chromatographi coupl mass spectrometri analyst detect vast arrai compound potenti order thousand accur analysi reli heavili skill limit pool experienc analyst utilis semi automat techniqu analys dataset make outcom subject focu current laboratori data analysi system refin exist approach work repres paradigm shift achiev appli data mine techniqu tackl problem techniqu compel efficaci preprocess method essenti applic area object evalu paper present preliminari result data mine framework predict concentr petroleum compound water sampl experi demonstr framework produc model suffici accuraci measur term root geoffrei holm dale fletcher peter reutemann eib frank squar error correl coeffici offer potenti significantli reduc time spent analyst task accuraci machin learn model versu hand craft expert system credit score case studi cluster document wikipedia dbase concept represent abstract paper show wikipedia semant knowledg exploit document cluster creat concept base document represent map term phrase document articl concept wikipedia develop similar measur evalu semant related concept set document test concept base represent similar measur standard text document dataset empir result show optim perform approach improv relat techniqu accuraci machin learn model versu hand craft expert system credit score case studi revisit multipl dinstanc learn embed instanc select mile recent propos multipl instanc mi classif algorithm appli singl instanc base learner proposition version mi data origin author singl instanc base learner algorithm norm svm present empir studi investig efficaci altern base learner mile compar mile mi algorithm result show boost decis stump case provid classif accuraci norm svm base learner mile mile competit perform compar mi learner identifi simpler proposition method requir shorter train time retain mile strong classif perform dataset test discrimin class dclass versu multi dclass classif abstract applic requir abil identifi data anomal respect target group observ sens belong previous unseen attack class approach kind verif problem class classif learn descript target class concern base sole data class target class train time standard multi class class classif exploit neg data infer descript target class paper assum scenario hold investig condit multi class class na ve bay classifi prefer class model aim identifi exampl attack class end identifi perform fair comparison techniqu concern present adapt standard cross valid main contribut paper base experiment result obtain show condit group techniqu prefer main find multi class class classif prefer class classif suffici larg number target class dclass classif combin densiti class probabl estim abstract class classif import applic outlier novelti detect commonli tackl densiti estim techniqu adapt standard classif algorithm problem carv decis boundari describ locat target data paper investig simpl method class classif combin applic densiti estim form refer distribut induct standard model class probabl estim method refer distribut gener artifici data emploi form artifici class conjunct target class artifici class basi standard class learn problem explain densiti function refer distribut combin class probabl estim obtain form adjust estim densiti function target class uci dataset data typist recognit problem show combin model consist densiti estim class probabl estim improv compon techniqu class classif compar method class classif support vector machin addit regress appli larg dscale collabor filter problem abstract public netflix competit put spotlight applic domain collabor filter spark interest machin learn algorithm appli sort problem demand natur netflix data lead interest ingeni modif standard learn method effici speed basic method appli approach netflix problem stand neighborhood base method latent factor model base singular decomposit ensembl consist variat techniqu paper investig applic forward stage wise addit model netflix problem regress scheme base learner ensembl weight simpl linear regressor mean cluster interpret tool multivari regress context experiment result show method produc competit result dclass classif combin densiti class probabl estim abstract class classif import applic outlier novelti detect commonli tackl densiti estim techniqu adapt standard classif algorithm problem carv decis boundari describ locat target data paper investig simpl method class classif combin applic densiti estim form refer distribut induct standard model class probabl estim method refer distribut gener artifici data emploi form artifici class conjunct target class artifici class basi standard class learn problem explain densiti function refer distribut combin class probabl estim obtain form adjust estim densiti function target class uci dataset data typist recognit problem show combin model consist densiti estim class probabl estim improv compon techniqu class classif compar method class classif support vector machin cluster document activ learn wikipedia wikipedia appli background knowledg base text mine problem attempt made util document cluster paper propos exploit semant knowledg wikipedia cluster enabl automat group document similar theme cluster intrins unsupervis recent research shown incorpor supervis improv cluster perform limit supervis provid approach present paper appli supervis activ learn util wikipedia creat concept base represent text document concept wikipedia articl exploit semant related wikipedia concept find pair wise instanc level constraint supervis cluster guid cluster direct constraint test approach standard text document dataset empir result show basic document represent strategi yield compar perform previou attempt ad constraint improv cluster perform empir comparison exact nearest neighbour algorithm abstract nearest neighbour search nn problem practic import number field involv find point call queri point set point nearest queri initi incept problem great number algorithm techniqu propos solut remain case propos algorithm compar wide varieti dataset research attempt fill gap extent present detail empir comparison promin data structur exact nn kd tree metric tree cover tree result suggest gener gain metric tree cover tree kd tree standard nn problem improv bag input smear naiv bay text classif unbalanc class abstract multinomi naiv bay mnb popular method document classif due comput effici good predict perform recent establish predict perform improv data transform paper present transform design combat potenti problem applic mnb unbalanc dataset propos correct adjust attribut prior correct implement data normal step show significantli improv area roc curv show modifi version mnb close relat simpl centroid base classifi compar method empir ensembl balanc nest dichotomi multi dclass problem abstract system nest dichotomi hierarch decomposit multi class problem class classproblem repres tree structur ensembl randomlygener nest dichotomi proven effect approach multi class learn problem sampl tree give tree equal probabl mean depth tree limit number class unbalanc tree neg affect runtim paper investig approach build balanc nest dichotomi class balanc nest dichotomi data balanc nest dichotomi evalu ensembl set decis tree base model show approach reduc runtim effect accuraci problem class investig effect cach model build ensembl nest dichotomi unsupervis discret tree dbase densiti estim speed logist model tree induct abstract logist model tree shown accur compact classifi greatest disadvantag comput complex induc logist regress model tree address issu aic criterion crossvalid prevent overfit model addit weight trim heurist produc signific speedup compar train time accuraci induct process origin dataset show train time decreas classif accuraci diminish slightli ensembl balanc nest dichotomi multi dclass problem abstract system nest dichotomi hierarch decomposit multi class problem class classproblem repres tree structur ensembl randomlygener nest dichotomi proven effect approach multi class learn problem sampl tree give tree equal probabl mean depth tree limit number class unbalanc tree neg affect runtim paper investig approach build balanc nest dichotomi class balanc nest dichotomi data balanc nest dichotomi evalu ensembl set decis tree base model show approach reduc runtim effect accuraci problem class investig effect cach model build ensembl nest dichotomi weka machin learn workbench data mine weka machin learn workbench data mine gene select microarrai data cancer classif machin learn approach logist model tree tree induct method linear model popular techniqu supervis learn task predict nomin class numer valu predict numer quantiti work combin scheme model tree tree linear regress function leav paper present algorithm adapt idea classif problem logist regress linear regress stagewis fit process construct logist regress model select relev attribut data natur show approach build logist regress model leav increment refin construct higher level tree compar perform algorithm state art learn scheme benchmark uci dataset show produc accur compact classifi toolbox learn relat data proposit multi dinstanc learner classif evalu output confid dbase associ rule mine abstract associ rule mine data mine techniqu reveal interest relationship databas exist approach emploi paramet search interest rule fact larg number rule make difficult compar output confid base associ rule miner paper explor classif perform metric evalu output previou work form classifi associ rule focuss accur classif concentr properti result classifi basi compar confid base associ rule learner present experiment result uci dataset show qualiti small rule set gener apriori improv predict apriori algorithm show cba standard method classif associ rule gener inferior standard rule learner run time size rule set ensembl nest dichotomi multi dclass problem nest dichotomi standard statist techniqu tackl polytom classif problem logist regress repres binari tree recurs split multi class classif task system dichotomi provid statist sound appli class learn algorithm multi class problem assum algorithm gener class probabl estim candid tree problem standard approach choic tree base domain knowledg practic altern treat system nest dichotomi equal form ensembl classifi base assumpt show approach produc accur classif appli logist regress directli multi class problem result show ensembl nest dichotomi produc accur classifi pairwis classif techniqu compar result logist regress compar error correct output code prefer logist regress compar case addit benefit gener class probabl estim good generalpurpos method appli binari classifi multi class problem logist regress boost label bag instanc paper upgrad linear logist regress boost multi instanc data consist label bag instanc connect predict individu instanc bag level probabl estim simpl averag maxim likelihood bag level word assum instanc contribut equal independ bag label present empir result artifici data gener underli gener model assum show algorithm produc competit result musk benchmark dataset evalu replic signific test compar learn algorithm abstract empir research learn algorithm classif task gener requir signific test qualiti test typic judg type error test differ type ii error differ paper argu replic test import test low replic outcom strongli depend random partit data perform present empir measur replic compar perform popular test realist set involv standard learn algorithm benchmark dataset base result give recommend test data mine bioinformat weka predict librari congress classif librari congress subject head paper address problem automat assign librari congress classi cation lcc work set librari congress subject head lcsh lcc organ tree root node hierarchi compris topic leaf node correspond special topic area de ned describ procedur resourc identi ed lcsh automat place resourc lcc hierarchi procedur machin learn techniqu train data larg librari catalog learn classi cation model map set lcsh node lcc tree present empir result techniqu show accuraci independ collect lcsh lcc pair logist model tree tree induct method linear model popular techniqu supervis learn task predict nomin class numer valu predict numer quantiti work combin scheme model tree tree linear regress function leav paper present algorithm adapt idea classif problem logist regress linear regress stagewis fit process construct logist regress model select relev attribut data natur show approach build logist regress model leav increment refin construct higher level tree compar perform algorithm state art learn scheme benchmark uci dataset show produc accur compact classifi dlevel learn method gener multi dinstanc problem abstract tradit multi instanc mi learn singl posit instanc bag produc posit class label learner bag class label depend label instanc bag explicitli inform solv learn task paper investig gener view mi problem simpl assumpt longer hold assum interact instanc bag determin class label level learn method type problem transform mi bag singl meta instanc learn standard proposit method meta instanc region instanc space cover instanc bag result artifici realworld data show level classif approach suit gener mi problem local weight naiv bay simplic naiv bay classifi surpris machin learn research exhibit good perform varieti learn problem encourag result research look overcom naiv bay primari weak attribut independ improv perform algorithm paper present local weight version naiv bay relax independ assumpt learn local model predict time experiment result show local weight naiv bay rare degrad accuraci compar standard naiv bay case improv accuraci dramat main advantag method compar techniqu enhanc naiv bay conceptu comput simplic visual class probabl estim induc classi er make accur predict futur data drive forc research induct learn import user gain inform model produc power induct learn algorithm gener black box represent model make virtual imposs gain insight learn paper present techniqu user understand classi er make predict provid dimension visual class probabl estim requir classi er gener class probabl practic algorithm modi ed end local weight naiv bay simplic naiv bay classifi surpris machin learn research exhibit good perform varieti learn problem encourag result research look overcom naiv bay primari weak attribut independ improv perform algorithm paper present local weight version naiv bay relax independ assumpt learn local model predict time experiment result show local weight naiv bay rare degrad accuraci compar standard naiv bay case improv accuraci dramat main advantag method compar techniqu enhanc naiv bay conceptu comput simplic race committe larg dataset paper propos method gener classifi larg dataset build committe simpl base classifi standard boost algorithm permit process larg dataset underli base learn algorithm effici multiclass altern decis tree simpl approach ordin classif abstract machin learn method classif problem commonli assum class valu unord practic applic class valu exhibit natur order learn grade standard approach ordin classif convert class numer quantiti appli regress learner transform data translat output back discret class post process step disadvantag method appli conjunct regress scheme paper present simpl method enabl standard classif algorithm make order inform class attribut appli conjunct decis tree learner show outperform naiv approach treat class valu unord set compar special purpos algorithm ordin classif method advantag appli modif underli learn scheme determin progress glaucoma visual field interact machin learn let user build classifi standard procedur build classier fulli autom process data prepar domain expert contrast interact machin learn engag user gener classier oer natur integr background knowledg model stage long interact tool design support cient eectiv commun paper show techniqu empow user creat model compet classier built state art learn algorithm demonstr user user domain expert construct good classier learn algorithm simpl dimension visual interfac experi demonstr surprisingli success hing domain attribut support good predict user gener accur classier domain high order attrib text categor compress model text categor assign natur languag text predefin categori base content grow import volum inform internet continu overwhelm predefin categori impli supervis learn approach categor classifi articl bottom dup proposition paper present method proposition work bottom data driven manner tailor biochem databas exampl descript chemic compound method gener frequent fragment linearli connect atom user speci length preliminari experi domain carcinogen predict show bottomup proposition promis approach featur construct relat data introduct approach handl relat learn problem transform proposit problem subsequ appli proposit learn algorithm transform requir construct featur captur relat properti learn exampl featur construct model construct decoupl transform relat learn problem proposit call proposition naiv bay regress technic note data mine practic machin learn tool techniqu java implement abstract paper present discoveri rule challeng encount video index rule consid knowledg guidelin develop index tool fuzzi decis tree extract rule base color proport kei frame extract singl video new experiment result comparison data mine tool present kea practic automat keyphras extract keyphras provid semant metadata summar character document paper describ kea algorithm automat extract keyphras text kea identifi candid keyphras lexic method calcul featur valu candid machin learn algorithm predict candid good keyphras machin learn scheme build predict model train document keyphras model find keyphras document larg test corpu evalu kea effect term author assign keyphras correctli identifi system simpl robust gnu gener public licens paper instruct gener rule set model tree abstract model tree decis tree linear model leaf node recent emerg accur method numer predict produc understand model decis list order set rule potenti compact understand tree counterpart present algorithm induc simpl accur decis list model tree model tree built repeatedli rule select iter method produc rule set accur smaller model tree construct entir dataset experiment result heurist attempt find compromis rule accuraci rule coverag report show method produc compar accur smaller rule set commerci state art rule learn system cubist make global discret appli learn algorithm dataset practition global discret numer attribut algorithm handl numer attribut directli prior discret essenti prior discret acceler induct produc simpler accur classifi gener global discret deni learn algorithm chanc take advantag order inform implicit numer attribut simpl transform discret data preserv inform form learner show compar discret data directli transform significantli increas accuraci decis tree built decis list built part decis tabl built wrapper method benchmark dataset significantli reduc size result classifi simpl techniqu make global discret tool data preprocess domain dspecif keyphras extract keyphras import mean document summar cluster topic search small minor document author assign keyphras manual assign keyphras exist document labori highli desir autom keyphras extract process paper show simpl procedur keyphras extract base naiv bay learn scheme perform compar state art explain procedur perform boost automat tailor extract process document collect hand result larg collect technic report comput scienc show qualiti extract keyphras improv significantli domain specif inform exploit kea practic automat keyphras extract keyphras provid semant metadata summar character document paper describ kea algorithm automat extract keyphras text kea identifi candid keyphras lexic method calcul featur valu candid machin learn algorithm predict candid good keyphras machin learn scheme build predict model train document keyphras model find keyphras document larg test corpu evalu kea effect term author assign keyphras correctli identifi system simpl robust gnu gener public licens paper instruct gener accur rule set global optim domin scheme rule learn ripper oper stage induc initi rule set refin complex optim stage discard adjust ripper individu rule make work contrast paper show good rule set learn rule time global optim present algorithm infer rule repeatedli gener partial decis tree combin major paradigm rule gener creat rule decis tree separ conquer rule learn techniqu algorithm straightforward eleg experi standard dataset show produc rule set accur similar size gener accur ripper oper effici avoid postprocess suffer extrem slow perform patholog set method critic permut test attribut select decis tree techniqu attribut select decis tree bias attribut valu ad hoc solut problem appear machin learn literatur statist test exist associ prespecifi signific level provid wellfound basi address problem statist test comput chi squar distribut valid approxim actual distribut larg sampl case patent hold leav decis tree except class permut test describ permut test appli problem choos test explor give stage method appli select attribut decis tree result practic dataset compar favor method adopt pre prune strategi introduct statist test provid set model tree classif model tree type decis tree linear regress function leav form basi recent success techniqu predict continu numer valu appli classif problem emploi standard method transform classi cation problem problem function approxim surprisingli simpl transform model tree induc base quinlan gener accur classifi state art decis tree learner attribut numer 