Credit Card Fraud Detection via Kernel-Based Supervised Hashing

Zhenchuan Li, Guanjun Liu, Shuo Wang, Shiyang Xuan, Changjun Jiang

Abstract: Although credit card fraud detection has been studied for many years, these detection models cannot effectively help financial experts handle fraud alerts since they only predict a risk degree for a transaction but are unable to provide any information to explain why the transaction is of this risk degree. This paper presents a Kernel-based Supervised Hashing (KSH) model to detect credit card fraud. KSH is based on the idea of approximate nearest neighbor that can provide the most similar existing fraud samples for a transaction when the transaction is predicted to be fraud. These similar samples can help experts analyze the transaction, improve the detection accuracy and lower the disturbing rate. Additionally, KSH is very suitable for the large and high-dimension dataset. To the best of our knowledge, it is the first time that KSH is used to detect credit card fraud, and our experiments on a real large transaction dataset illustrate its advantages and effectiveness.