Ccf crypto

Comment

Author: Admin | 2025-04-28

Fraudulentsource.The goal of detectionsoftware is to classify eachtransaction as legal orfraudulent. The types of errors that can occur in this classification areclassification of a fraudulent transaction as legal (false negative); andclassificationofalegaltransactionasfraudulent(falsepositive).Probabilityofdetection=PD=Pr(classifyintoH1│H1istrue)orProbabilityoffalsenegative=1–PDProbabilityoffalsepositive=PF=Pr(classifyintoH1│H0istrue)LetthenumericalvaluesforthenormalandfraudulenttransactionsfollowexponentialdistributionswithparametersλNandλF,λN>λFrespectively.TheprobabilityofdetectionPDandprobabilityoffalsepositivePFasttxFDFFedxePλλλ−∞−==∫ttxNFNNedxePλλλ−∞−==∫ThusPDcanbeexpressedasafunctionofPFasrFDPP =wherer=λF/λNisbetween0and1.Trees [18] stated that the quality profile of most detection software ischaracterizedbyacurvethatrelatesitsPDandPF known asthe receiver operatingcharacteristic curve (ROC). ROC is a function that summarizes the possibleperformancesofadetector.Itvisualizesthetrade-offbetweenfalsealarmratesanddetectionrates,thusfacilitatingthechoiceofadecisionfunctions.ThiswasthereforeusedintheperformanceanalysisofCCFandotherdetectionwatchsystems.The effectiveness of this detection software is measured in terms of theclassification errors, which consist of systemdetectionrate and false alarm rate.The data used in the application were collected from a Nigerian bank, whichconsistoftransactiondata madeperday during theobserved period, thatis,inamonth. The collection was done according to the two types of bank operationsinvestigated. The aim is to identify frauds within each of the categories byidentifyingflaggedtransactions usingtheCCF detection watchapproach.DetailsofthetestdatasetsarelistedinTable1.Table1.SummaryoftheTwoDataSubsetsUsedtoTesteachoftheOperationsofCCFDetectionModel.Operation Transaction Fraudulent ProportionoffraudulentWithdrawal 10,650 5 0.47%Deposit 8,102 2 0.24%Total 18,752 7 0.37% DataMiningApplicationinCreditCardFraudDetectionSystem317JournalofEngineeringScienceandTechnologyJune2011,Vol.6(3)The performance analyses of the respective detection algorithms are car riedoutusingMATLABsoftwarepackageandtheresultscomparedwiththecollecteddataareasshowninFigs.4and5.Fig.4.ROCforWithdrawalFraudDetection.Fig.5.ROCforDepositFraudDetection.Figures4and5showthatthemodelresults comparedsatisfactorilywellwiththecollecteddataresultsforthetwotransactionsexaminedinthispaper.Inthe comparisonof CCF detection watch performance with twootherfrauddetection models using the ROC curve, it wa sno ted that the results from CCFdetection using neural network was accurate and reliable. The two differentcommercial products, quadratic discriminates analysis (QDA) and logisticregression (LOGIT) wereselected to test the feasibilityofusing neural networktools for the purpose of CCF detection watch. The performance anal ysis of theCCF detection watch model (deposit transaction) is compared with these twocommercialpackagesandtheresultisshowninFig.5.FromFig.6,theROCcurveforCCFdetectionwatchdetectsover95%offraudcaseswithoutcausingfalsealarms.ThisisfollowedbytheROCcurveforlogisticregression with 75% detection with no false alarms, the quadratic discriminant 318F.N.OgwuelekaJournalofEngineeringScienceandTechnologyJune2011,Vol.6(3)analysis with 60% detection. This shows that the performance of CCF detectionwatchisinagreementwithotherdetectionsoftware,butperformsbetter.Fig.6.ComparisonofCCFDetectionWatchwithotherFraudDetectionSystemROCforDepositFraudDetection.Using two software products enabled this work to illustrate different userinterfaces available, alternative to neural networks mod els, design for decision-making, and the performance metrics of different models. Comparison ofperformanceofneuralnetworkmodelwithtraditionalstatisticalmodelsincreasedthe confidence in the ability of CCFdetection watc hin successful modelling ofcreditcardfraudsinthebankingindustry.In the comparison of the performance evaluation of the two- stage clustermodelandthe four-stagecluster model,thetwo-stagecredit card frauddetectionmodel works as a binary classifier that has two choices, “fraud” or “legal”. Itworks by producing a score.  The score is a measure of the confidence of theclassifierthataparticular transactionislegal or fraud. To decide whichitis,thescoreiscomparedtoathresholdthenitisdeemedtobelegal.Ifitisgreater,thenit is fraud. This type of model consists of a component and five user interfacesandtends to classify most legaltransactions as fraudulent. Thefour-stage creditcardfrauddetection modeldesignedconsistsof twocomponentswithseven userinterfaces.ThetwocomponentsareArtificialNeuralNetwork(ANN)componentand the Rule Based (RB) component. It was developed using four classes ofcluster-low,high,riskyandhighrisk.IntheSOMNNengine,thedatasetdescription,numberofdatapointsin thedataset,andthenumberofclustersareentered.Thedatapoint’sentrywascreatedandfilled.Theentriesaresorted inascending order under the cluster list. Whenthetrain/generateclustersareselected,the clusterlabelbecomesreadyforfillingdependingonthenumberofclustersenteredandthenstoredonthedatabase.Thedatabase was stored in Microsoft Access table and was also used to determinewhena card transactionwastobe processed,blocked,unblocked,orthe alertsetoff.After each transaction, the datapoint entry and clusters madeareprocessedbytheSOMNNengineandsentintothedatabase.Thishelpsthedetectionenginetoknowwhenanydataentryislegitimateorfraudulent,andthe reasonis givenimmediately after the alert. The transaction will not be processed, but will be DataMiningApplicationinCreditCardFraudDetectionSystem319JournalofEngineeringScienceandTechnologyJune2011,Vol.6(3)storedinthedatabase.TheSOMNNsetupreportdisplaysthenumberofclusters,thenamesoftheclustersandthelistofclustertransactions.The database was meant to run at the background of the existing bankingsoftware and be getting its data from real-time banking transaction, checkingwhether the transaction is legitimate (and so will be processed) or fraudulent(transaction will not be processed and the alert will be let off with reasonsdisplayed). The detection software i si nterfaced with the source data. MicrosoftAccess was used to create and manage the database. Each of the data tablescreatedinthisstudysuchastransactiontables,accountopentables,clustertables,legitimate and fraudulent tables, is stored in a separate file. In the cluster nametable, four cluster names were used. Since the

Add Comment