8.4 The Fraud Profile

Advances were made during the 1990s, enabling banks to be more effective in detecting credit-card fraud via detection systems consisting of a detailed transaction-history database coupled with a sophisticated, real-time neural network and rule-based scoring system. A purchasing profile is built on each user, which is quite small, like a string of DNA chromosomes, using years of data. The detection system determines the patterns of a cardholder's shopping habits. These patterns include such information as frequency of purchases, average purchases, location of purchases, and other transactional factors. All of this information is very compact; for example, a consumer's string can be 8 (number of purchases), $45 (average purchase price), 94### (location zip code of purchases), etc. Eventually, all of this transaction history is used to construct a knowledge base about each consumer.

Before a transaction can be authorized, the real-time detection system attempts to pattern match the transaction with the knowledge base of the cardholder's historical transactions. In addition, the system will attempt to pattern match the transaction with knowledge bases of known fraudulent transactions. Pattern recognition aims to classify data based on either a prior knowledge or on statistical information extracted from the patterns. The patterns to be classified are usually groups of measurements or observations. In this case, the pattern recognition attempts to put the transaction into either a legitimate or a fraudulent category.