Wavelet-Based Approach for Automatic Seizure Detection Using EEG Signals

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 09 Issue: 08 | Aug 2022 www.irjet.net p-ISSN: 2395-0072

Wavelet-Based Approach for Automatic Seizure Detection Using EEG Signals

1Associate Professor, ECE Department, Geetanjali Institute of Technical Studies, Udaipur, Rajasthan, India 2ECE Department, Bennett University, Greater Noida, India 3Professor, Electrical Department, Geetanjali Institute of Technical Studies, Udaipur, Rajasthan, India ***

Abstract - People with epilepsy frequently experience seizures, which reduce their quality of life. When electroencephalograph (EEG) recordings of these individuals are categorized precisely between seizure-free and seizure-based segments, it is possible to look into previous seizures and forecast upcoming ones. Modeling an EEG signal can help with the extraction of discriminative characteristics. In this study, a wavelet-based approach is used to break down EEG signals into detailed and approximate coefficients up to the fourth level of decomposition. For the purpose of distinguishing between normal EEG data and signals recorded from epileptic patients, several statistical features have been derived from the wavelet coefficients and fed to various classifiers. The simulation results showed that the suggested model, when applied to the neurology and sleep centre EEG database, New Delhi, attained the maximum classification accuracy of 100 % between healthy and epileptic EEG signals.

When compared to cutting-edge techniques detailed in the literature, the proposed model exhibits more accuracy.

Key Words: (Crossvalidation,Electroencephalogram(EEG),Epilepticseizure,Statisticalfeatures,Waveletcoefficients,support vectormachine.

1. INTRODUCTION

Oneofthemostprevalentneurologicalconditions,epilepsyaffectsabout50millionpeopleglobally[1].Themostvulnerable groupsincludeolderpeopleandchildren,whoseprevalenceratesrangefrom0.7to1.0percent,aswellasthoseindividuals withconcomitantconditions[2].Discrimination,misunderstanding,anddepressionarecommonexperiencesforpeoplewith epilepsy.Meanwhile,thisdisorderisdangeroussinceitcanleadtodeathifapersonwithepilepsyengagesina dangerous activity[3][4].Epilepsydiagnosisandtreatmentdecisionsarebothcustomized.Neurologistswillidentifysomeoneashaving epilepsyiftheyhaveanepilepticseizureandsubsequentseizureswithinthefollowing24hours.Patientsmaypotentiallyhave epilepsyifthereisachancetheywillexperienceanotherseizureaftertwounprovokedonesduringthenext8–10years[5]. Seizurepreventioncanbecarriedouttoavertbraindamageifaseizureisdiagnosedearlyenough.Duringanepilepticseizure, theelectricalbehaviorofthebrainsignaldiffersfromnormalbrainactivityintermsofshape,amplitudes,andfrequency[6].If epilepsyisidentifiedearlyonandtreatedwithmedicationandsurgery,itisprojectedthattwo-thirdsofthoseaffectedwilllive seizure-free lives. The researchers have used a variety of neuro-imaging methods to precisely identify seizures [7]. Electroencephalogram(EEG)isoneoftheforemosttoolsinneuroscienceforassessingbrainabnormalities,mostlyforseizure detection[8][9][10].Theaccuratediagnosisofepilepticseizuresiscarriedoutbyskilledandtrainedneurologistsusing continuousmonitoringandinterpretationoftheEEGrecordings.Thisisatime-consuming,expensive,andcomplextaskthat couldresultinanincorrectdiagnosisbecausetrainedprofessionalsareoverworked[8].Asaresult,researchershavetherefore madenumerousattemptstoautomaticallyidentifyepilepticseizures.

Sameeretal.in[11]employedshort-timeFouriertransform(STFT)toextractdeltarhythm,fromthetime-frequencyanalysis ofEEGsignal.Fourstatisticalfeaturesnamelykurtosis,mean,variance,andskewnesshavebeencomputedfromthedeltaband oftheEEGsignal.Usingarandomforest(RF)classifier,theproposedalgorithmachievedaclassificationaccuracyof97.40% whilediscriminatingbetweenpeoplesufferingfromepilepsyandhealthypeople.In [12],theauthorscomputedepileptic seizuredensityasafeaturethatisfedtok-nearestneighbor(kNN)toevaluatetheperformanceoftheproposedmethodology. UsingneurologyandsleepEEGdataset,thepresentstudyachievedgoodaccuracyof99%whenclassifyingbetweenpre-ictal andictalEEGsignals.Theauthorsin[13]demonstratedahybridapproachbasedonmulti-scaleradialbasicfunctionandthe Fishervectortechniqueforinvestigatingthehigh-resolutiontime-frequencyestimationtoanalyzethedynamicbehaviorofthe non-stationaryEEGsignals.Sharmaetal.in[14]usedanovelmodelbasedonanorthogonalwaveletfilterbank(OWFB)for discriminationofictalandnon-ictalEEGsignalsusingtheBONNEEGdatabaseandneurologyandsleepcenterEEGdatabase. Thesuggestedmethod, whenusedwithten-foldcross-validation,wasabletodistinguishbetween pre-ictal and ictal EEG signalswithaclassificationaccuracyof98%andbetweeninter-ictalandictalEEGsignalswithaclassificationaccuracyof100

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 09 Issue: 08 | Aug 2022 www.irjet.net p-ISSN: 2395-0072

%.Astochasticdifferentialequations-basedapproachfortheprecisecategorizationofseizureandhealthyEEGsignalswas reportedbyTajmirriahietal.in[15].Thesuggestedmodelachievedthemaximumclassificationaccuracyof99.1%forthe neurologyandsleepcentredatabasebetweenhealthyandictalEEGsignalsusingasupportvectormachine(SVM). Carvalhoet al.in[16]explainedfiveadaptivedecompositiontechniquesfortheanalysisofnon-stationaryandnon-linearEEGsignals.Itis revealedthatamongthesemethods,variationalmodedecompositionandempiricalmodedecompositionbasedapproach showedsuperiorresultsintermsofclassificationaccuracy.Hadiyosoetal.in [17]proposedtwofeaturesnamelyrelative waveletenergyandwaveletentropyfortheaccuratedetectionofanepilepticseizure.TheEEGsignalsaredecomposedinto five frequency rhythmsfromwhich two featureshave been extractedfrom eachof thesefrequency bands.Usingan SVM classifier,thesimulationresultshowedthehighestclassificationaccuracyof96%forinter-ictalvsictalEEGsignals.Adiscrete cosinetransform(DCT)basedfilterbankwassuggestedbyGuptaetal.in[18]tobreakdowntheEEGsignalsintofivedifferent brainrhythms.ForthebinaryclassificationofEEGsegments,twofeatures theHurstexponentandautoregressivemoving average aregeneratedfromtheserhythmsandprovidedasinputstotheSVMclassifier.Theefficacyofthepresentstudyis assessedintermsofevaluationmetricsusingtwopubliclyavailableEEGdatabases.

Thestructureofthispaperisasfollows:Thedatabaseusedinthisstudy,thewavelet-baseddecompositionoftheEEGsignals, thefeatureextractionandselectionprocesseshavebeendiscussedinSectionII,andtheclassifiersemployedarecoveredin SectionIII.SectionIVpresentsthesimulationresultsanddiscussion.Thefinalportioncontainstheconclusion.

2. PROPOSED METHOD

Inthiswork,thedatasetincludessegmentedEEGrecordingsoftenepilepticpatientsfromtheNeurologyandSleepCentre, HauzKhas,NewDelhi[19].AmplificationequipmentfromGrassTelefactorcalledtheCometAS40wasusedtorecordthe signals.WhileobtainingtheEEGrecords,gold-platedscalpEEGelectrodesareplacedinaccordancewiththe10-20electrode placementschemes.Abandpassfilterwithafrequencyrangeof0.5to70Hzwasusedtofilteralloftherecordingsafterthey hadbeensampledat200Hz.EachEEGsegmentisclassifiedbysegmentationintooneofthreecategories:pre-ictal,interictal, orictal.Theentiredatasetisdividedinto150segments(50foreachcategory).EachtimeseriesEEGrecordhasalengthof5.12 seconds,or1024samples.This datasetconsists offifty MATfilesorganizedintotheictal,inter-ictal,andpre-ictal named folders.Inthispaper,apre-ictalfolderisnamedasSetA,aninter-ictalfolderasSetB,andanictalfolderasSetC,whichare showninFigure1.AflowchartoftheproposedmethodisdepictedinFigure2.

Fig -1:AsampleofSetA,SetB,andSetCaspreictal,interictalandictalEEGsignals

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 09 Issue: 08 | Aug 2022 www.irjet.net p-ISSN: 2395-0072

Fig -2:Aflowchartofproposedmethod

Fig -3:PreictalEEGsignalwithitsdetailedandapproximatecoefficients

2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page383

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 09 Issue: 08 | Aug 2022 www.irjet.net p-ISSN: 2395-0072

Fig -4:InterictalEEGsignalwithitsdetailedandapproximatecoefficients

Fig -5:IctalEEGsignalwithitsdetailedandapproximatecoefficients

2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page384

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 09 Issue: 08 | Aug 2022 www.irjet.net p-ISSN: 2395-0072

2.1 DISCRETE WAVELET TRANSFORM

Wavelet transform is one of the most promising, efficient, and appealing tools for time-frequency representation of nonstationarysignalssuchasECG,EEG,andothers.WavelettransformisprimarilyusedasareplacementforFouriertransform(FT) andshort-timeFouriertransform(STFT),as itaddresses the shortcomingsof bothand provides both time andfrequency componentsatanyinstant,revealingthehiddenfeaturesoftheoriginalsignalmuchmoreeffectively.Highfrequenciesarebetter resolvedinthetimedomainofthewavelettransform,whereaslowfrequenciesarebetterresolvedinthefrequencydomain. WavelettransformsarecommonlyusedonEEGsignalsforthreemainpurposes:denoisingofsignal,featureextraction,and signalcompression[20][21].Wavelettransformhasthepropertytochangeitsfinitewindowlengthandlocationdepending upontranslationandscalingparameters(b,k)whichareexpressedas:

wherethemotherwaveletisrepresentedbythefunctionψb,k,whichisscaledbythefactorkandtranslatedbythefactorof b.Whenthemotherwaveletismultipliedwithanysignalandintegratedoveralltimes,itsresultisfurthermultipliedwith 1/√kforthenormalizationpurposeandyieldwavelettransform.Continuouswavelettransformistheresultofconvolution ofthesignalx(t)andwaveletfunctionψ(t)andisexpressedas:

Intheliterature,ithasbeenobservedthatneitherSTFTnorCWTpossessestheirpracticalimplementationusinganalytical equations and integrals, thus there is a need to use them in discrete format. It has been investigated that even after the discretizationofCWT,computationbecomespossiblebyusingcomputers,however,thisisnotanoriginaldiscretewavelet transform(DWT)asitsimplyworksonthesamplingofwaveletinCWT,whichleadstolotsofredundantinformationwhile reconstructing a signal and ultimately increases computational time. On the other hand, actual DWT is a highly efficient transform thatsignificantlyreduces thecomputationtime andcan reconstructsignals more efficiently.DWT isfrequently employed as a filter bank (comprising low pass and high pass filters) for signal segmentation. Detailed and approximate coefficientshavebeencomputeduptothefourthlevelofdecomposition.Thedetailedcoefficientsrepresenthigh-frequency componentsoftheEEGsignalwhileapproximatecoefficientsrepresentlow-frequencycomponents.Inthiswork,theHaar waveletisemployedasabasicwaveletfunction.

2.2 FEATURE EXTRACTION

Inthispaper,tenstatisticalfeatures,namelymean,median,range,meanabsolutedeviation,medianabsolutedeviation, standarddeviation,L1norm,L2norm,maximum,andminimumvalueshavebeencomputedfromwaveletcoefficientsknown asdetailedandapproximatecoefficients.Thesewaveletcoefficientsareextractedfrompre-ictal,inter-ictal,andictalEEG signals.AbriefdescriptionofthesestatisticalfeaturesisdepictedinTable1.

2.3 FEATURE SELECTION USING WILCOXON RANK-SUM TEST

AftercomputingthestatisticalfeaturesfromthewaveletcoefficientsofeachofthethreeclassesofEEGsignals,thenextstepis toacquireasubsetoftherelevantfeaturesusingstatisticaltechniques.TheWilcoxonrank-sumtestisusedinthisstudytopick significantfeaturesusingtheMATLABstatisticaltoolbox,withp-valueandz-scoreassignedata95percentsignificancelevel. Thisapproachisusedtodeterminehowsimilarthepopulationlocationsare(mediansequaltozeroornot).

Thenullhypothesisarguesthattwopopulationshavecomparabledistributionfunctions,whereasthealternativehypothesis statesthattheyaredistinctintermsofmedians.Anyfeaturewithap-valuelessthan0.05areconsideredarelevantfeaturethat canbeinputintothemachinelearningalgorithmtoimproveclassificationaccuracy.Othertraitswithp-valuesgreaterthan 0.05,ontheotherhand,areunimportantandcanberemoved.

3 CLASSIFICATION

Inthisstudy,threeclassifiersnamelySVM,kNN,andensemblesubspacekNNclassifiershavebeenemployedtoassessthe efficacyoftheproposedmodel.Abriefdescriptionoftheseclassifiersisasfollow:

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 09 Issue: 08 | Aug 2022 www.irjet.net p-ISSN: 2395-0072

3.1 SUPPORT VECTOR MACHINE (SVM)

SVMclassifierissolelybuiltusingstatisticallearningtheoryin1995byCortesandVapnik.Itisoneofthebest-supervised classifiers which is used for binary classification. Also, now-a-days, researchers are using an SVM classifier for pattern classification purposes due to its regularization parameter to avoid over fitting, selection of optimal kernel, and convex optimization[22].Thisclassifiercompletelyworksonthearchitectureofthekernelfunction.Thiskernelfunctionisthemost prominenttool,whichhelpsinselectingthemostefficienthyperplanetoseparatetrainingsamplesofbinaryclassifierwithout errorandalsoresultsindistancemaximizationfromboththeclassestothisseparatinghyperplaneasdiscussedabove[23] SomeofthefamouskernelfunctionsofanSVMclassifierarethepolynomialkernel,linearkernel,andradialbasisfunctions (RBF).IntheSVMclassifier,thelineardiscriminantfunctionisq(x)=py+bandtheseparatinghyperplanecanbeexpressedby theequationpy+b=0.

3.2 k-NEAREST NEIGHBOR (kNN)

Thisclassificationapproachisconsidered oneofthesimplestapproachesasitfollowsa non-parametricapproach.Itcan classifyagivendatapointaccordingtoviewingitsmajorityofneighborhoodpoints.ThekNNalgorithmmainlyfollowsatwostepapproach.Theprimarystepistofindthenumberofnearestneighbors,whereasinthesecondstepclassificationofadata pointintoaparticularclasswouldbedonebyreferringtotheprimarystep.TheEuclideandistancetechniqueisusedinthis studytodeterminetheclosestneighbor.

3.3 ENSEMBLE SUBSPACE kNN

Bymixingthepredictionsfromvariousmodels,ensemblelearningisabroadMetaapproachtomachinelearningthataimsto improve predictive performance. In this study, random subspace ensembles are used to increase the k-nearest neighbor classifiers'accuracy.Subspaceensembleshavethebenefitsofutilizinglessspaceandhandlingmissingvaluesthanensembles withallpredictors.

4. RESULT AND DISCUSSION

Inthisstudy,binaryproblemshavebeencategorizedusingthethreeperformancemeasuresofsensitivity(Sen),specificity (Spec),andaccuracy(Acc)inordertoassesstheeffectivenessofthesuggestedmethod.

Thesemetricsaredefinedas:

Table -1: Listofstatisticalfeaturesextractedfromwaveletcoefficients

Featureindex Featurename Description

1

Mean Itrepresentstheaveragevalueofthegivensignal.

2 Median Itrepresentsthemiddlevaluesofthegivendatapoints. 3

Range Itrepresentsthedifferencebetweenmaximumand minimumvalue

4

Meanabsolutedeviation Itindicatestheaveragedistancebetweeneachobservation andmeanvalueofdata.

5

Medianabsolutedeviation

Itisameasureofvariabilityinthegivensignal

6 Standarddeviation Itrepresentsthedeviationofdatapointsfromthemean value

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 09 Issue: 08 | Aug 2022 www.irjet.net p-ISSN: 2395-0072

7

L1Norm

Sumofabsolutevaluesofdatapoints

8 L2Norm ItrepresentstheEuclideandistanceofdatapointfromthe origin

9 Maximum Maximumvalueofthegivensignal

10 Minimum Minimumvalueofthegivensignal

Table -2: AcomparisonofEEGsignalclassificationperformanceofproposedmethodwiththeexistingtechniques

Authorsandyear Techniqueused Featuresused EvaluationMetrics

Guptaetal.[18], 2021 Discretecosine transform Hurstexponent Accuracy=96.5%

Lietal.[13],2019 Hybridapproach Sub-bandsfeatures Accuracy=99.3%

Sharmaetal.[14], 2018 OWFB Entropybased features Accuracy=96.2%

Hadiyosoetal.[24], 2021 Wavelettransform Relativeenergyand entropybased features

Accuracy=96%

Sameeretal.[11], 2020 STFT Kurtosis,mean, variance,skewness Accuracy=98%

Proposedmethod Discretewavelet transform Statisticalfeatures Accuracy=100%

Here,theterms TP, TN,FP,andFNstandfortherespectivetermstruepositiverate,truenegativerate,false-positiverate, andfalse-negativerate.TodistinguishbetweenhealthyandepilepticEEGsignalsisthemaingoalofthecurrentinvestigation. Thecategorizationchallengehasbeenincludedinthecurrentworkusingthepubliclyaccessibledataset. Bysegmentingthe completebandwidthofEEGsignalsintowaveletcoefficientsusingdiscretewavelettransform,thedecompositionofhealthyand epilepticEEGsignalsisobtained.A10-foldcross-validationmethodwasusedinthecurrentstudytodeterminehowwellthe machinelearningmodelperformed. Eachofthetenequallysizedchunksofthecompletedatasampleischosenasatrainingset ataspecificpointintime. Thefirstcomponentisusedtotestthemodelduringthefirstiteration,whiletheremainingsections areusedfortraining. Thesecondcomponentisusedfortestinginthefollowingiteration,andtheremainingsectionsareusedto trainthemodelandsoforth. Thisprocedureisrepeateduntilatestingsetisusedforeachofthetensections.

Inthisstudy,pre-ictal,inter-ictal,andictalEEGsignals,recordedfromtensubjectscontaining1024samplesoverdurationof 5.12seconds,areconsideredtoevaluatetheresults.Discretewaveletbasedapproachisemployedtodecomposethreeclassesof EEG signals into various detailed and approximate coefficients as shown in Figures 3, 4, and 5 respectively. A total of ten statisticalfeatureshavebeenextractedfromdetailedandapproximatecoefficientsofpreictal,interictalandictalEEGsignals. Thesefeaturesarepassedtovarietyofclassifierssuchask-nearestneighbor(kNN),SVM,andensemble(subspacekNN)forthe classificationofbinaryproblem.ThenumerousresearchesthataddresstheissueofclassifyingepilepticandnormalEEGare compared in Table 2. When distinguishing between inter-ictal and ictal EEG signals, the suggested technique achieved exceptionalclassificationaccuracyof100percentwhencomparedtootherstrategiesdiscussedintheliterature.Theproposed modelalsoshowedbetterepilepticdetectionof97.2%whencomparingbetweenpre-ictalandictaltypeofEEGsignals.

Additionally, Figures 6,7,and 8 depict theconfusion matrix for SVM,kNN,and ensemble subspacekNN types ofmachine learningmodels.Itisconcludedfromthesefiguresthatthesuggestedmodelshowedgoodperformancebyutilizinganytypeof supervisedmachinelearningmodels.Bycomparingpredictedandactualclasses,itisawell-knowndescriptionusedtoimagine theclassifier'seffectiveness.Itshowstheprecisenumberofinstancesthatwerecorrectlyandincorrectlycategorized

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 09 Issue: 08 | Aug 2022 www.irjet.net p-ISSN: 2395-0072

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

5. CONCLUSIONS

ThisstudypresentsaneffectivemethodforclassifyingEEGsignalsaseithernormalorthoserecordedduringepilepticseizure activity.Usingdiscretewavelettransform,tenstatistical featureshavebeencomputedfrom the waveletcoefficients.The significanceoftheextractedfeaturesisperformedbyapplyingWilcoxonranksumtest.Anyfeaturehavingp-valuelessthan 0.005isdiscardedat95%significantlevel.ThesefeaturesarethensuppliedtovarietyofclassifierssuchaskNN,SVM,and ensemble to assess the success rate of the proposed method. The proposed study, which outperformed the previous approaches,used10-foldcross-validationtoachieveaclassificationaccuracyof100%whendifferentiatingbetweeninter-ictal andictalEEGdata.Thistechniquemaybeusedinthefuturetodiagnoseavarietyofbrainillnesses.

REFERENCES

[1] World Health Organization. (2019). Epilepsy. [Online]. Available: https://www.who.int/en/news-room/fact sheets/detail/epilepsy,accessedon28-Aug-2019.

[2] Kuroda,N.EpilepsyandCOVID-19:Associationsandimportantconsiderations.EpilepsyandBehavior,108:107122,2020

[3] Moshé,S.L.,Perucca,E.,Ryvlin,P.,Tomson,T.Epilepsy:Newadvances.TheLancet,385(9971):884-898,2015.

[4] Wijayanto,I.,Rizal,A.,Humairani,A. SeizuredetectionbasedonEEGsignalsusingKatzfractalandSVMclassifiers.In2019 5thInternationalConferenceonScienceinInformationTechnology(ICSITech),Yogyakarta,Indonesia,pp.78-82,2019.

[5] Falco-Walter,J.J.,Scheffer,I.E.,Fisher,R.S.Thenewdefinitionandclassificationofseizuresandepilepsy.EpilepsyResearch, 139:73-79,2018.

[6] Tripathi,D.,Agrawal,N.Epilepticseizuredetectionusingempiricalmodedecompositionbasedfuzzyentropyandsupport vectormachine.In:HwangS.,TanS.,BienF.(eds)ProceedingsoftheSixthInternationalConferenceonGreenandHuman InformationTechnology.ICGHIT2018.LectureNotesinElectricalEngineering,vol.502.Springer,Singapore,2019.

[7] AnyanwuC,MotamediG.K.Diagnosisandsurgicaltreatmentofdrug-resistantepilepsy.JBrainSci8(4):1–20,2018.

[8] AdeliH,ZhouZ,DadmehrcN.AnalysisofEEGrecordsinanepilepticpatientusingwavelettransform.JNeurosciMethods 123(1):69–87,2003.

[9] Liu A, Hahn JS, Heldt GP, Coen R.W. Detection of neonatal seizures through computerized EEG analysis. ElectroencephalogramClinNeurophysiol82(1):30–37,1992.

[10] BoashashB,MesbahM,ColditzP.B. TimefrequencydetectionofEEGabnormalities,Time-FrequencySignalAnalysisand Processing:AComprehensiveReference,pp663–670,2003.

[11] Sameer, M. and Gupta, B. Detection of Epileptic Seizures based on alpha band statistical features. Wireless Personal Communications,pp.1-16,2020.

[12] Darjani, N. and Omranpour, H. Phase space elliptic density feature for epileptic EEG signals classification using metaheuristicoptimizationmethod.Knowledge-BasedSystems,205,p.106276,2020.

[13] Li,Y.,Cui,W.G.,Huang,H.,Guo,Y.Z.,Li,K.andTan,T. EpilepticseizuredetectioninEEGsignalsusingsparsemultiscale radialbasisfunctionnetworksandtheFishervectorapproach.Knowledge-BasedSystems,164,pp.96-106,2019.

[14] Sharma,M.,Bhurane,A.A.andAcharya,U.R. MMSFL-OWFB:Anovelclassoforthogonalwaveletfiltersforepilepticseizure detection.Knowledge-BasedSystems,160,pp.265-277,2018.

[15] Tajmirriahi,M.andAmini,Z. Modelingofseizureandseizure-freeEEGsignalsbasedonstochasticdifferentialequations. Chaos,SolitonsandFractals,150,p.111104,2021.

[16] Carvalho,V.R.,Moraes,M.F.,Braga,A.P.andMendes,E.M. Evaluatingfivedifferentadaptivedecompositionmethodsfor EEGsignalseizuredetectionandclassification.BiomedicalSignalProcessingandControl,62,p.102073,2020.

[17] Hadiyoso, S., Irawati, I.D. and Rizal,A. Epileptic electroencephalogram classification using relative wavelet sub-band energyandwaveletentropy.InternationalJournalofEngineering,34(1),pp.75-81,2021.

Volume: 09 Issue: 08 | Aug 2022 www.irjet.net p-ISSN: 2395-0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page389

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 09 Issue: 08 | Aug 2022 www.irjet.net p-ISSN: 2395-0072

[18] Gupta,A.,Singh,P.andKarlekar,M. Anovelsignalmodelingapproachforclassificationofseizureandseizure-freeEEG signals.IEEETransactionsonNeuralSystemsandRehabilitationEngineering,26(5),pp.925-935,2018.

[19] Swami, P., Panigrahi, B. K., Nara, S., Bhatia, M., Gandhi, T. K. EEG epilepsy datasets. http://dx.doi.org/10.13140/RG.2.2.14280.32006,2016.

[20] Alyasseri,Z.A.A.,Khader,A.T.,Al-Betar,M.A.,andAbualigah,L.M. ECGsignaldenoisingusing$\beta$-hillclimbing algorithmandwavelettransform,In8thInternationalConferenceonInformationTechnology,pages1-7,2017.

[21] Kumar,H.,Pai,S. P.,Vijay,G.,andRao,R. Wavelet transform for bearingcondition monitoringandfaultdiagnosis:A review,InternationalJournalofCOMADEM,17(1):9-23,2014.

[22] Subasi,A.andGursoy,.I. EEGsignalclassificationusingPCA,ICA,LDAandsupportvectormachines,ExpertSyst.Appl.,vol. 37,no.12,pp.8659–8666,2010.

[23] Christianini, N., and Shawe-Taylor, J. C. An introduction to support vector machines and other kernel-based learning methods.Cambridge:CambridgeUniversityPress,2000.

[24] S.Hadiyoso,I.D.Irawati,andA.Rizal, Epilepticencephalogramclassificationusingrelativewaveletsub-bandenergyand waveletentropy,Internationaljournalofengineering,vol.34,no.1,pp.75-81,2021.

2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal |