?url_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Adc&rft.title=Algoritmos+de+aprendizaje+autom%C3%A1tico+para+detecci%C3%B3n+de+fraudes+con+tarjetas+de+cr%C3%A9dito%3A+An%C3%A1lisis+y+comparativa&rft.creator=Calvo+P%C3%A9rez%2C+Ismael&rft.contributor=G%C3%B3mez+Canaval%2C+Sandra&rft.contributor=Arroyo+Montoro%2C+Fernando&rft.subject=Computer+Science&rft.description=El+pago+con+tarjetas+de+cr%C3%A9dito+es+una+pr%C3%A1ctica+cada+vez+m%C3%A1s+habitual+en+la+sociedad+actual.+Por+desgracia%2C+esta+actividad+se+ve+constantemente+amenazada+por+posibles+fraudes+en+cualquier+tipo+de+sus+transacciones.+Estas+vulnerabilidades+acarrean+consecuencias+negativas+para+la+industria+financiera+y+los+usuarios.+La+miner%C3%ADa+de+datos+ha+desempe%C3%B1ado+un+papel+destacado+en+la+detecci%C3%B3n+del+fraude+con+tarjetas+de+cr%C3%A9dito.+En+concreto%2C+la+detecci%C3%B3n+de+transacciones+fraudulentas+puede+enfocarse+como+un+problema+de+aprendizaje+autom%C3%A1tico%2C+el+cual+se+afronta+como+un+reto+debido+a+dos+razones+principales%3A+en+primer+lugar%2C+los+perfiles+de+los+comportamientos+normales+y+fraudulentos+cambian+constantemente+y%2C+en+segundo+lugar%2C+los+conjuntos+de+datos+de+fraude+con+tarjetas+de+cr%C3%A9dito+est%C3%A1n+muy+sesgados+(distribuci%C3%B3n+de+clases+desequilibrada).+El+rendimiento+de+la+detecci%C3%B3n+del+fraude+en+las+transacciones+de+tarjetas+de+cr%C3%A9dito+se+ve+afectado+en+gran+medida+por+el+enfoque+de+muestreo+del+conjunto+de+datos%2C+la+selecci%C3%B3n+de+variables+y+la(s)+t%C3%A9cnica(s)+de+detecci%C3%B3n+utilizadas.+Este+Proyecto+de+Fin+de+Grado+aborda+el+estudio+de+las+t%C3%A9cnicas+existentes+para+tratar+con+conjuntos+de+datos+desequilibrados+y+la+detecci%C3%B3n+de+fraudes+con+tarjetas+de+cr%C3%A9dito+mediante+algoritmos+de+Aprendizaje+Autom%C3%A1tico.+Para+ello%2C+se+lleva+a+cabo+un+an%C3%A1lisis+comparativo+de+las+t%C3%A9cnicas+de+submuestreo+(aleatorio)+y+sobremuestreo+(SMOTE)+de+datos+y%2C+tras+balancear+la+muestra%2C+se+implementan+cuatro+algoritmos+clasificadores+cl%C3%A1sicos%2C+a+saber%3A+Regresi%C3%B3n+Log%C3%ADstica%2C+k-vecinos+m%C3%A1s+pr%C3%B3ximos%2C+M%C3%A1quina+de+Soporte+Vectorial+(SVM)+y+%C3%81rbol+de+Decisiones.+Por+%C3%BAltimo%2C+se+desarrollan+dos+redes+neuronales+profundas+(una+para+cada+m%C3%A9todo+de+balanceo)+cuyos+resultados+se+comparan+con+el+algoritmo+que+mayor+rendimiento+haya+ofrecido+en+sus+predicciones.+Las+pruebas+experimentales+de+este+Proyecto+se+han+basado+en+un+conjunto+de+datos+(dataset)+p%C3%BAblico%2C+obtenido+a+trav%C3%A9s+de+la+plataforma+Kaggle+y+a+partir+del+cual+se+ha+desarrollado+un+%E2%80%98notebook%E2%80%99+implementado+en+Python.%0D%0AAbstract%3A%0D%0ACredit+card+payments+are+an+increasingly+common+practice+in+today%E2%80%99s+society.+Unfortunately%2C+this+activity+is+constantly+under+threat+from+credit+card+fraud%2C+which+has+negative+consequences+both+for+the+financial+industry+and+the+final+user.+Data+mining+has+played+a+prominent+role+in+the+detection+of+the+credit+card+frauds+using+online+transactions.+In+particular%2C+fraudulent+transaction+detection+can+be+identified+as+a+machine+learning+problem%2C+which+becomes+challenging+due+to+two+main+reasons%3A+firstly%2C+the+profiles+of+normal+and+fraudulent+behaviours+in+the+transactions+are+constantly+changing+and%2C+secondly%2C+credit+card+fraud+datasets+are+highly+skewed+(unbalanced+class+distribution).+The+performance+of+fraud+detection+in+credit+card+transactions+is+largely+affected+by+the+sampling+approach+of+the+dataset%2C+the+selection+of+variables+and+the+detection+technique(s)+used.+This+Final+Degree+Project+adressess+the+study+of+existing+techniques+for+dealing+with+unbalanced+data+sets+within+the+detection+of+credit+card+frauds+context+using+Machine+Learning+algorithms.+For+this+purpose%2C+a+comparative+analysis+of+the+techniques+of+random+undersampling+and+oversampling+(SMOTE)+of+data+is+carried+out+and%2C+after+balancing+the+sample%2C+four+classic+classifier+algorithms+are+implemented%2C+namely%3A+Logistic+Regression%2C+k-nearest+neighbours%2C+Support+Vector+Machine+(SVM)+and+Decision+Trees.+Finally%2C+two+deep+neural+networks+are+developed+(one+for+each+balancing+method)+whose+results+are+compared+with+the+algorithm+that+has+offered+the+best+performance+in+its+predictions.+The+experimental+tests+of+this+Project+have+been+based+on+a+public+dataset+from+the+Kaggle+platform%2C+from+which+a+notebook+implemented+in+Python+has+been+developed.&rft.publisher=E.T.S.I+de+Sistemas+Inform%C3%83%C2%A1ticos+(UPM)&rft.rights=https%3A%2F%2Fcreativecommons.org%2Flicenses%2Fby-nc-nd%2F3.0%2Fes%2F&rft.date=2021-07&rft.type=info%3Aeu-repo%2Fsemantics%2FbachelorThesis&rft.type=Final+Project&rft.type=PeerReviewed&rft.format=application%2Fpdf&rft.language=spa&rft.rights=info%3Aeu-repo%2Fsemantics%2FopenAccess&rft.identifier=https%3A%2F%2Foa.upm.es%2F67976%2F