模糊规则库与人工神经网络对酵母数据分类性能的比较研究(IJITCS-V7-N5-6)

I.J.InformationTechnologyandComputerScience,2015,05,40-47PublishedOnlineApril2015inMECS()DOI:10.5815/ijitcs.2015.05.06Copyright©2015MECSI.J.InformationTechnologyandComputerScience,2015,05,40-47AComparativeStudyonthePerformanceofFuzzyRuleBaseandArtificialNeuralNetworktowardsClassificationofYeastDataShrayasiDattaDepartmentofInformationTechnology,JalpaiguriGovernmentEngineeringCollege,Jalpaiguri,WestBengal,IndiaEmail:[e-mail]fInformationTechnology,KalyaniGovernmentEngineeringCollege,Kalyani,Nadia,WestBengal,India.Email:[e-mail]—Classificationofyeastdataplaysanimportantroleintheformationofmedicinesandinvariouschemicalcomponents.Ifthetypeofyeastcanberecognizedattheprimarystagebasedontheinitialcharacteristicsofit,alotoftechnicalprocedurecanbeavoidedinthepreparationofchemicalandmedicalproducts.Inthispaper,theperformancetwoclassifyingmethodologiesnamelyartificialneuralnetworkandfuzzyrulebasehasbeencompared,fortheclassificationofproteins.Theobjectiveofthisworkistoclassifytheproteinusingtheselectedclassifyingmethodologyintotheirrespectivecellularlocalizationsitesbasedontheiraminoacidsequences.TheyeastdatasethasbeenchosenfromUCImachinelearningrepositorywhichhasbeenusedforthispurpose.Theresultshaveshownthattheclassificationusingartificialneuralnetworkgivesbetterpredictionthanthatoffuzzyrulebaseonthebasisofaverageerror.IndexTerms—ProteinLocalization,Classification,NeuralNetwork,FuzzyRuleBase,YeastDatasetI.INTRODUCTIONAcellusuallycontainsapproximate1billion(or109)proteinmolecules[1],[2].Theseproteinmoleculesresideinvariouscompartmentsofacellwhichusuallycalled―proteinsubcellularlocations‖.Theinformationaboutthesesubcellularlocationshelpstoknowthefunctionsofthecellandthebiologicalprocessexecutedbythecells.Thisinformationalsohasbeenusedfortheidentificationofdrugtargets([3],[4]).Determiningthesubcellularlocalizationofaproteinbyconductingbio-chemicalexperimentsisalaboriousandtimeconsumingtask.Butwiththedevelopmentofmachinelearningtechniques[5]incomputerscience,togetherwithanincreaseddatasetofproteinsofknownlocalization,fastandaccuratelocalizationpredictionsformanyorganismshavebeendonesuccessfully.Thisisduetothenatureofmachinelearningapproaches,whichperformedwellindomainswherethereisavastcollectionofdatabutwithalittletheory–whichperfectlydescribesthesituationinbioinformatics[5].Amongvariousprokaryoticandeukaryoticorganisms,yeastisimportantbecausethesearewidelyusedinmedicineandinfoodtechnologyfield.Biologicalstructureofyeasthasalsosnatchedtheattentionofresearchersformanyyearsbecauseoftheirsimilaritywithhumancell.Forpredictingthesubcellularlocalizationofyeastprotein,thefirstapproachhasbeendevelopedbyKanehisaandNakai([6],[7]).HortonandNakai[8]haveproposedaprobabilisticmodelwhereexperthasidentifiedthosefeatureswhichlearnitsparametersfromasetoftrainingdata.Theauthorsalsohaveimplementedandtestedthreemachinelearningtechniquesnamelyk-nearestneighboralgorithm,binarydecisiontree,naïveBayesclassifierinyeastdatasetandE.Colidataset[9].PerformanceofthesethreetechniqueswiththeProbabilisticmethod[8]hasalsobeencomparedandithasbeenshownthattheperformanceofk-nearestneighboralgorithmisbetteramongthesefour.ChenY.[10]hasimplementedthreemachinelearningclassificationalgorithms:decisiontree,perceptron,two-layerfeedforwardnetworkforpredictingsubcellularlocalizationsiteofaproteinofyeastandE.Colidataset.Anditisconcludedthatthreetechniqueshassimilarperformancemeasureforthistwodataset.Qasim,R,Begum,K.Jahan,N.Ashrafi,T.Idris,S.Rahman,R.M.[11],haveproposedanautomatedfuzzyinterferencesystemforproteinsubcellularlocalization.BoJin,YuchunTang,Yan-QingZhang,Chung-DarLuandIreneWeber[12],haveproposedanddesignedSVMwithfuzzyhybridkernelbasedonTSKfuzzymodelandhaveshowedthatfuzzyhybridkernelhasachievedbetterperformanceinSVMclassification.Predictionofproteinsubcellularlocalizationworkhasbeendonein([13]-[16]).Outofthese,supportvectormachinetechniqueshavebeenusedin([13]-[15]).Alotofdecentworkalsohasbeendoneonwebserverdesignforsubcellularprediction([17]-[20]).AlgorithmbasedonFuzzyrulebasetechniqueisproposedinheartdiseaseandinpacketdeliverytime([21]-[23]).Classificationisdonewithsomewidelyusedmachinelearningtechniques,like,KNN,multilayeredfeedforwardneuralnetwork,SVMetc.([6]-[16]),butmostofAComparativeStudyonthePerformanceof41FuzzyRuleBaseandArtificialNeuralNetworktowardsClassificationofYeastDataCopyright©2015MECSI.J.InformationTechnologyandComputerScience,2015,05,40-47theworkisbasedonsomecomparisonwithotherdatasets,likeE.Coli,fungietc.Theymostlyhaveconcentratedonthealgorithm,i.e.whichalgorithmisbestsuitedforclassificationtaskofmedicaldatasets.Butforaparticulardataset,whichalgorithmismostefficienthasnotbeenchecked.Andthatiswhytheworkdescribedinthispaperhasbeentaken.Here,apopularandveryimportantproteinsubcellularlocalizationdataset,yeast,hasbeentakenforclassification,andmultilayeredfeedforwardneuralnetworkandfuzzyrulebasetechniquehasbeenusedandcomparedforclassificationtask.YeastdatasetfromUC