Otkrivanje financijskih prijevara pomoću strojnog učenja

master's thesis

• Title (english): Detection of Financial Frauds Using Machine Learning
• Author: Robert Jambrečić
• Mentor: Marin Šilić (mentor)
• Committee member: Marin Šilić (predsjednik povjerenstva)
• Committee member: Klemo Vladimir (član povjerenstva)
• Committee member: Goran Delač (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2020-07-07, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Za rješavanje problema otkrivanja financijskih prijevara isprobane su metode nadziranog i nenadziranog strojnog učenja. Ako se koristi neuravnotežen skup za učenje, tradicionalni nenadzirani modeli postižu bolje rezultate od tradicionalnih nadziranih modela. Isprobane su različite arhitekture autoenkodera kao metoda predobrade podataka, ali bez doprinosa kvaliteti sustava. Za uspješno rješavanje zadatka, ključno je odrediti način rješavanja problema neuravnoteženog skupa podataka. Tradicionalni... More nadzirani modeli strojnog učenja postižu najbolje rezultate ako se na skupu za učenje provede SMOTE uravnotežavanje. S druge strane, potpuno povezana neuronska mreža i konvolucijska neuronska mreža najbolje rezultate postižu uz neuravnoteženi skup za učenje. Pritom treba težine postaviti tako da model kod učenja obraća veću pažnju na manjinsku klasu. Less

Abstract (english)

For detecting financial frauds, supervised and unsupervised machine learning methods have been tested. If an unbalanced training set is being used, traditional unsupervised models achieve better results than traditional supervised models. Different autoencoder arhitectures are tried as data preprocessing method, but without contributing to the quality of the system.
For task solving success, it is crucial to determine class balancing method.
Traditional machine learning supervised models... More achieve the best results if the training set is balanced using the SMOTE algorithm. On the other hand, fully connected neural network and convolutional neural network achieve the best results, while using the unbalanced training set, but with setting the weights that the model uses to pay more attention to the minority class while training. Less

• Keywords: nadzirano; nenadzirano; uravnotežavanje; SMOTE; strojno; duboko; učenje; slučajna; šuma; neuronska; konvolucijska; mreža; autoenkoder
• Keywords (english): supervised; unsupervised; balancing; SMOTE; machine; deep; learning; random; forest; neural; convolutional; network; autoencoder
• Language: croatian
• URN:NBN: urn:nbn:hr:168:647115
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-01-13 12:30:28