Two-stage credit card fraud detection using sequence alignment

Abstract

A phenomenal growth in the number of credit card transactions, especially for on-line purchases, has also led to a substantial rise in fraudulent activities. Implementation of efficient fraud detection systems has thus become imperative for all credit card companies in order to minimize their losses. In real life, fraudulent transactions could be interspersed with genuine transactions and simple pattern matching techniques are not often sufficient to detect the fraudulent transactions efficiently. In this paper, we propose a hybrid approach in which anomaly detection and misuse detection models are combined. Sequence alignment is used to determine similarity of an incoming sequence of transactions to both a genuine card holder’s sequence as well as to sequences generated by a validated fraud model. The scores from these two stages are combined to determine if a transaction is genuine or not. We use stochastic models for studying the performance of the system.