Biometric Recognition System based on Analysis and Classification of Physiological Signals

Abstract

This dissertationexplorestheinnovativedomainofbiometricrecognitionsys- tems, focusingontheanalysisandclassificationofphysiologicalsignalsasamethod to improvepersonalidentificationandauthenticationprocesses.Thediscussion delvesintovariousphysiologicalsignalsfusionintoourdevelopedbiometricframe- work,suchaselectrocardiography(ECG),ImpedanceCardiography(ICG)and ContinuousBloodPressure(BP)signalshighlightingtheirsignificanceinbiomet- ric applications.ByemployingadvancedmachinelearningtechniquesandANN, signal processingalgorithmsandablationstudy,thecurrentstudydemonstrates howthesesignalsprovideuniquecharacteristicpatterntoeachindividualtailored with multitasksoftheproposedbiometricsystem.Furthermore,thisdissertation addresses thechallengesfacedinreal-worldimplementations,includingdatapri- vacyconcerns,andtheneedforrobustclassificationmodels.Resultsindicatethe Fine Gaussian-SVMmodelachievedan88.14%accuracyduringtraining,witha recall of95.09%,precisionof94.33%,andaKappacoefficientof87.7%.Inthe test set,FG-SVMdemonstrated93.33%accuracy,balancedrecallandprecisionof 93.33%, andaKappacoefficientof92.9%.TheBi-layeredANNmodelexhibited superiortrainingperformance,attaining93.3%accuracy,94.56%recall,93.17% precision, andaKappacoefficientof93.1%.Notably,inthetestset,Bi-layered ANN achievedperfectaccuracy,recall,precision,andKappacoefficientof100%. The presentedfindingsenrichthedataset,aimtocontributetothegrowingbody of knowledgeinbiometrictechnology,showcasingthepotentialofbloodpressure signal analysisasacornerstonefornext-generationbiometricrecognitionsystems. This researchoffersvaluableinsightsforacademicandhealthcaresectorswhich enhance operationalefficiencybyimprovingpatientsatisfactionthroughmitigate misidentificationofpatients,whilealsominimizingcosts,medicalerrors,andpre- ventingfraudofstakeholdersinterestedinthefutureofbiometricauthentication solutions.