SVDD-based financial fraud detection methodthrough respective learnings of normal/abnormal behaviors

Abstract

This thesis proposes a method to detect financial fraudby dividing users' financial transactions into a normal area and an abnormal area,using SVDD and train the areas as such fraud evolves in terms of complexity. The existing financial industry detects electronic financial frauds using FDS, but its false positive rate is high enough to require additional authentications of user information. It causes customers inconveniences and does not always detect those sophisticated financial frauds. In order to resolve the aforementioned issues, this study proposes a method to detect such potential frauds by profiling user financial transaction data including user activities, device information, andtransaction patterns and vectorizing them into a normal area and an abnormal area using SVDD.