Mining relationships among interval-based events for classification[1]

《Mining relationships among interval-based events for classification[1]》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、MiningRelationshipsAmongInterval-basedEventsforClassificationDhavalPatelWynneHsuMongLiLeeSchoolofComputingNationalUniversityofSingaporeSingapore{dhaval,whsu,leeml}@comp.nus.edu.sgABSTRACTtemporalpatternsfrominstantaneousevents,thatis,eventswithnoduration.Thisassumptionallowsthedi

2、scoveredExistingtemporalpatternminingassumesthateventsdopatterntobesimplifiedtoanorderedsequenceofevents,nothaveanyduration.However,eventsinmanyrealworldsuchas“fever→stomachache→vomit”.However,suchapplicationshavedurations,andtherelationshipsamongsequentialpatternsareinadequateto

3、expressthecomplextheseeventsareoftencomplex.Theserelationshipsaremod-temporalrelationshipsindomainssuchasmedical,multi-eledusingahierarchicalrepresentationthatextendsAllen’smedia,meteorologyandfinancewheretheevents’durationsintervalalgebra.However,thisrepresentationislossyastheco

4、uldplayanimportantrole.exactrelationshipsamongtheeventscannotbefullyrecov-Forexample,ithasbeenobservedthatinmanydiabeticered.Inthispaper,weaugmentthehierarchicalrepresenta-1patients,thepresenceofhyperglycemiaoverlapswiththetionwithadditionalinformationtoachievealosslessrepre-2ab

5、senceofglycosuria.Thisinsighthasledtothedevelop-sentation.AnefficientalgorithmcalledIEMinerisdesignedmentofeffectivediabetictestingkits.Inthecaseofdenguetodiscoverfrequenttemporalpatternsfrominterval-basedfever,knowingthattherewillbeadecreaseofplateletcountsevents.Thealgorithmemplo

6、ystwooptimizationtechniquesonthethirddayaftertheonsetoffeverhasledtoabettertoreducethesearchspaceandremovenon-promisingcan-managementofthedisease.Clearly,thereisaneedforanef-didates.Fromthediscoveredtemporalpatterns,webuildficientminingalgorithmthatcandiscovercomplexrelation-anin

7、terval-basedclassifiercalledIEClassifiertodifferenti-shipsamongeventswithduration,alsoknownasinterval-atecloselyrelatedclasses.Experimentsonbothsyntheticbasedevents.Furthermore,thesediscoveredrelationshipsandrealworlddatasetsindicatetheefficiencyandscalabilitycouldbeusedtobuildaclass

8、ifierthatisabletodistinguishoftheproposedapproac